New energy generation power prediction method and system considering turning meteorological conditions

Abstract

The invention relates to the technical field of new energy power generation, in particular to a new energy power generation power prediction method and system considering turning meteorological conditions. Acquiring weather forecast and historical power data, performing feature extraction by using the Grubrum angle field data conversion and the efficient channel attention convolutional neural network, and constructing a turning weather recognition model; performing turning weather classification and historical similar day matching by utilizing a Gaussian mixture model and a mahalanobis distance according to a judgment result of the turning weather recognition model, and formulating a turning weather compensation strategy; constructing a power prediction fusion model based on a random forest model limit gradient boosting model and a long short-term memory neural network; and combining an output result of the power prediction fusion model with a compensation sequence of a turning weather compensation strategy to obtain a new energy power generation power prediction result. The problems of low efficiency and poor dynamic adaptability of turning weather feature extraction are solved, the prediction precision under the weather turning condition is improved, and the method has good prediction precision and generalization ability.

Description

Technical Field

[0001] The present invention relates to the technical field of renewable energy power generation, and in particular to a method and system for predicting renewable energy power generation taking into account transitional meteorological conditions. Background Art

[0002] New energy power forecasting technology can transform the uncertainty of new energy itself into the uncertainty of power forecasting error, greatly reducing the adverse effects of new energy randomness and volatility on the power system. Currently, the commonly used new energy power forecasting methods mainly include physical methods and data-driven methods. The physical method obtains the prediction results of meteorological elements and corresponding new energy power generation by establishing a mechanism model, while the data-driven method establishes the mapping relationship between relevant features and new energy power generation from a statistical perspective. Based on the rapid development of artificial intelligence and big data technology, data-driven methods have been widely used in new energy power forecasting.

[0003] With the continuous expansion of the scale of new energy installed capacity, the impact of transitional weather on the operation of new energy sites and power grids is increasing. The power generation of new energy is restricted by a variety of complex factors. The traditional single data-driven model cannot deeply explore the mapping relationship between various influencing factors and the power generation of new energy. At the same time, there is a lack of research on the identification of transitional weather and how it specifically affects the power generation of new energy. This limits the accuracy and adaptability of the prediction model, making it difficult to meet the requirements for power prediction accuracy. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to solve the problem of low efficiency in extracting transition weather features and poor dynamic adaptability, and how to reduce the impact of overfitting and effectively reduce the prediction error.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for predicting renewable energy power generation taking into account transitional meteorological conditions, which includes obtaining meteorological forecasts and historical power data, using Gram angle field data conversion and efficient channel attention convolutional neural network to perform feature extraction, and constructing a transitional weather recognition model;

[0008] According to the discrimination results of the transitional weather identification model, the Gaussian mixture model and Mahalanobis distance are used to classify the transitional weather and match it with similar historical days, and a transitional weather compensation strategy is formulated;

[0009] Construct a power prediction fusion model based on random forest model, extreme gradient boosting model and long short-term memory neural network;

[0010] The output result of the power prediction fusion model is combined with the compensation sequence of the transitional weather compensation strategy to obtain the new energy power generation power prediction result.

[0011] As a preferred solution of the method for predicting power generation of renewable energy taking into account transitional meteorological conditions of the present invention, wherein: the Gram angle field data conversion includes:

[0012] Normalize the input time series;

[0013] The normalized time series is mapped into a polar coordinate system, where the amplitude is encoded as the angular cosine and the time axis is encoded as the radius;

[0014] The Gram angle field matrix is ​​obtained by calculating the polar angle sum of the time series points after polar coordinate transformation.

[0015] As a preferred solution of the method for predicting power generation of renewable energy taking into account transitional meteorological conditions described in the present invention, the efficient channel attention convolutional neural network includes:

[0016] Input layer, used to input the image transformed by the Gram angular field matrix;

[0017] At least two convolutional layers, each followed by an efficient channel attention layer and a pooling layer;

[0018] The fully connected layer is used to flatten the output of the pooling layer and output the turning weather discrimination result.

[0019] As a preferred solution of the method for predicting power generation of renewable energy taking into account transitional meteorological conditions described in the present invention, the efficient channel attention layer includes:

[0020] Feature compression unit, used to perform global average pooling on input features;

[0021] Feature interaction unit, used to achieve local cross-channel interaction;

[0022] The weight generation unit is used to generate the weight coefficient of each channel.

[0023] As a preferred solution of the method for predicting power generation of renewable energy taking into account transitional meteorological conditions described in the present invention, the transitional weather compensation strategy includes:

[0024] The Gaussian mixture model is used to cluster the weather forecast and historical power data;

[0025] Match historical similar days in corresponding weather types based on Mahalanobis distance;

[0026] The actual power data of similar historical days is subtracted from the current predicted power data to obtain the power prediction compensation sequence.

[0027] As a preferred solution of the method for predicting power of renewable energy power generation taking into account transitional meteorological conditions described in the present invention, the process of constructing the power prediction fusion model includes:

[0028] The random forest model, extreme gradient boosting model and long short-term memory neural network model were established as basic prediction models respectively;

[0029] Calculate the prediction error of each basic prediction model on the validation set;

[0030] The weight of each model is obtained by normalizing the prediction error;

[0031] The prediction results of the three models are weighted averaged according to the weights to obtain the final prediction result.

[0032] As a preferred embodiment of the method for predicting renewable energy power generation taking into account transitional meteorological conditions described in the present invention, the renewable energy power generation prediction result is obtained by weighted combination of the output value of the power prediction fusion model and the compensation value of the transitional weather compensation strategy.

[0033] In a second aspect, an embodiment of the present invention provides a new energy power generation prediction system taking into account transitional meteorological conditions, which includes a feature extraction module, obtains meteorological forecasts and historical power data, uses Gram angle field data conversion and efficient channel attention convolutional neural network to extract features, and constructs a transitional weather recognition model;

[0034] A compensation strategy module, based on the discrimination result of the transitional weather identification model, uses a Gaussian mixture model and Mahalanobis distance to classify transitional weather and match it with similar historical days, and formulates a compensation strategy for transitional weather;

[0035] Prediction fusion module, which builds a power prediction fusion model based on random forest model, extreme gradient boosting model and long short-term memory neural network;

[0036] The result output module combines the output result of the power prediction fusion model with the compensation sequence of the transitional weather compensation strategy to obtain the new energy power generation power prediction result.

[0037] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the method for predicting new energy power generation power taking into account transitional meteorological conditions as described in the first aspect of the present invention are implemented.

[0038] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the method for predicting new energy power generation power taking into account transitional meteorological conditions as described in the first aspect of the present invention are implemented.

[0039] The beneficial effects of the present invention are as follows: constructing a transitional weather recognition model that combines the Gram angle field with the introduction of an efficient channel attention convolutional neural network to solve the problems of low efficiency in extracting transitional weather features and poor dynamic adaptability; using a Gaussian mixture model to achieve transitional weather condition classification, matching historical event similarities based on the Mahalanobis distance, and formulating a transitional weather compensation strategy to improve the prediction effect under transitional meteorological conditions; designing a power prediction fusion model that integrates the complementary advantages of multiple prediction models based on different principles, reduces the impact of overfitting, effectively reduces prediction errors, and improves the model's generalization ability. Compared with the prior art, the new energy power generation power prediction algorithm proposed by the present invention has good prediction accuracy and generalization characteristics under transitional meteorological conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 A flowchart of a method for predicting renewable energy power generation taking into account transitional meteorological conditions;

[0042] Figure 2 Computer equipment diagram for forecasting method of renewable energy power generation taking into account transitional meteorological conditions;

[0043] Figure 3 A schematic diagram of the Gram angle field matrix conversion for the renewable energy power generation prediction method taking into account transition meteorological conditions;

[0044] Figure 4 The structure diagram of the convolutional neural network combined with an efficient channel attention mechanism for the prediction method of renewable energy power generation taking into account transitional meteorological conditions;

[0045] Figure 5 A diagram of compensation effect of power forecast under transitional meteorological conditions for a new energy power generation power forecasting method taking transitional meteorological conditions into account;

[0046] Figure 6 Schematic diagram of the comparison curve of two prediction methods for renewable energy power generation taking into account transitional meteorological conditions. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0050] Example 1

[0051] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a method for predicting the power generation of new energy taking into account transition meteorological conditions, comprising:

[0052] S100: Obtain weather forecast and historical power data, use Gram angle field data conversion and efficient channel attention convolutional neural network for feature extraction, and build a transition weather recognition model;

[0053] Specifically, the meteorological forecast data that the present invention needs to obtain first include: basic meteorological elements such as temperature, humidity, wind speed, wind direction, air pressure, cloud cover, and dynamic meteorological elements such as the movement speed of the weather system and the development and evolution trend of cloud clusters. Historical power data include: historical power generation under different weather conditions, output characteristic curves, etc.

[0054] It should be noted that the transitional weather in the present invention generally refers to a situation in which the weather conditions change significantly within a relatively short period of time (such as 24 hours), such as sunny to cloudy, cloudy to overcast, overcast to light rain, etc. These weather transitions often lead to large fluctuations in the power generation of renewable energy, so special attention should be paid.

[0055] In an optional embodiment, the collection frequency of weather forecast data can be set to once every 15 minutes to ensure that weather change characteristics can be captured in a timely manner. The collection time span of historical power data is recommended to be no less than one year to cover the power generation characteristics under various weather conditions.

[0056] S101: Gram angle field data conversion includes:

[0057] Normalize the input time series;

[0058] The normalized time series is mapped into a polar coordinate system, where the amplitude is encoded as the angular cosine and the time axis is encoded as the radius;

[0059] The Gram angle field matrix is ​​obtained by calculating the polar angle sum of the time series points after polar coordinate transformation.

[0060] S102: Efficient Channel Attention Convolutional Neural Network includes:

[0061] Input layer, used to input the image transformed by the Gram angular field matrix;

[0062] At least two convolutional layers, each followed by an efficient channel attention layer and a pooling layer;

[0063] The fully connected layer is used to flatten the output of the pooling layer and output the turning weather discrimination result.

[0064] S103: Efficient channel attention layer includes:

[0065] Feature compression unit, used to perform global average pooling on input features;

[0066] Feature interaction unit, used to achieve local cross-channel interaction;

[0067] The weight generation unit is used to generate the weight coefficient of each channel.

[0068] Specifically, the special feature of the Gram angle field data conversion of the present invention is that it can not only retain the temporal characteristics of the time series, but also map the change trend of the data into the texture characteristics in the two-dimensional image. This conversion transforms the time series problem that was originally difficult to directly handle into an image recognition problem that is easier to handle.

[0069] It should be noted that the process of Gram angle field conversion is actually to construct a feature space, in which each point contains the state information of the original time series at that moment and the correlation information with other moments. This conversion method is particularly suitable for capturing the dynamic characteristics of the weather transition process.

[0070] In an optional embodiment, the innovation of the efficient channel attention layer is that by dynamically adjusting the weights of feature channels, the network can adaptively focus on the importance of different weather features. For example, in the process of sunny to cloudy, the weight of the cloud cover change feature will automatically increase; in the process of cloudy to rainy, the weight of the humidity and air pressure change features will increase accordingly.

[0071] S200: Based on the identification results of the transitional weather recognition model, the Gaussian mixture model and Mahalanobis distance are used to classify transitional weather and match it with similar historical days, and a transitional weather compensation strategy is formulated;

[0072] S201: Transitional weather compensation strategies include:

[0073] The Gaussian mixture model is used to cluster the weather forecast and historical power data;

[0074] Match historical similar days in corresponding weather types based on Mahalanobis distance;

[0075] The actual power data of similar historical days is subtracted from the current predicted power data to obtain the power prediction compensation sequence.

[0076] Specifically, the present invention adopts the Gaussian mixture model for weather classification because it can effectively handle the multimodal distribution characteristics of weather features and has the characteristics of self-adaptation. The use of Mahalanobis distance considers the correlation between features, making the matching of similar days more accurate.

[0077] It should be noted that the compensation strategy of the present invention not only considers weather similarity, but also considers temporal correlation. For example, even if the weather characteristics of two days are similar, if the time span is too large (such as winter and summer), its reference value will be reduced accordingly.

[0078] In an optional embodiment, the matching of similar historical days may adopt a multi-level screening strategy: first, a coarse screening is performed based on the weather type, then a fine screening is performed based on the Mahalanobis distance within the same type, and finally, a weight adjustment is performed considering the temporal proximity.

[0079] S300: Constructs a power prediction fusion model based on random forest model, extreme gradient boosting model and long short-term memory neural network;

[0080] S301: The process of constructing the power prediction fusion model includes:

[0081] The random forest model, extreme gradient boosting model and long short-term memory neural network model were established as basic prediction models respectively;

[0082] Calculate the prediction error of each basic prediction model on the validation set;

[0083] The weight of each model is obtained by normalizing the prediction error;

[0084] The prediction results of the three models are weighted averaged according to the weights to obtain the final prediction result.

[0085] Specifically, the present invention selects three models, random forest, extreme gradient boosting and long short-term memory network, because they have different advantages. Random forest is good at handling nonlinear relationships, extreme gradient boosting is insensitive to outliers, and long short-term memory network is good at capturing temporal dependencies.

[0086] It should be noted that the fusion of these three models is not a simple superposition of results, but a dynamic weight allocation based on their performance in different scenarios. The weight of the model will be adaptively adjusted as the prediction effect changes.

[0087] In an optional embodiment, the process of model fusion can be divided into two stages: first, offline weight allocation based on historical performance, and then online weight adjustment based on real-time prediction results. This dual weight mechanism can take into account both the stability and adaptability of the model.

[0088] S400: combining the output result of the power prediction fusion model with the compensation sequence of the transitional weather compensation strategy to obtain a new energy power generation power prediction result.

[0089] S401: The new energy power generation power prediction result is obtained by weighted combination of the output value of the power prediction fusion model and the compensation value of the transitional weather compensation strategy.

[0090] Specifically, when generating the final prediction result, the present invention not only considers the model fusion output, but also considers the compensation effect of the turning weather. The addition of the compensation value enables the prediction result to better reflect the power changes brought about by the weather turning.

[0091] It should be noted that the selection of compensation coefficients needs to balance the accuracy and stability of the forecast. Too large a compensation coefficient may lead to excessive fluctuations in the forecast results, while too small a compensation coefficient may not fully reflect the impact of weather changes.

[0092] In an optional embodiment, the compensation coefficient can be dynamically adjusted according to the severity of the weather change. For example, when a severe weather change is predicted, the compensation coefficient can be appropriately increased; when the weather change is relatively mild, the compensation coefficient can be reduced.

[0093] Furthermore, this embodiment also provides a new energy power generation prediction system taking into account transition meteorological conditions, including:

[0094] The feature extraction module obtains weather forecasts and historical power data, uses Gram angle field data conversion and efficient channel attention convolutional neural network to extract features, and builds a transition weather recognition model;

[0095] The compensation strategy module uses the Gaussian mixture model and Mahalanobis distance to classify the transitional weather and match it with similar historical days according to the judgment results of the transitional weather identification model, and formulates a compensation strategy for the transitional weather.

[0096] Prediction fusion module, which builds a power prediction fusion model based on random forest model, extreme gradient boosting model and long short-term memory neural network;

[0097] The result output module combines the output results of the power forecast fusion model with the compensation sequence of the transitional weather compensation strategy to obtain the new energy power generation power forecast results.

[0098] In summary, by introducing the transitional weather recognition model that combines Gram's angle field data conversion with an efficient channel attention convolutional neural network, the one-dimensional time series data is converted into a two-dimensional image with time information, and the feature weights are dynamically adjusted through the attention mechanism to achieve accurate extraction of transitional weather features. This step solves the technical problems of low feature extraction efficiency and poor dynamic adaptability of traditional methods during weather transitions, and the accuracy of transitional weather recognition is improved by 10%.

[0099] A compensation strategy for transitional weather is designed by combining a Gaussian mixture model with the Mahalanobis distance. First, the meteorological data are adaptively clustered using the Gaussian mixture model to avoid the limitations of the traditional fixed threshold classification method. Then, historically similar days are matched based on the Mahalanobis distance, taking into account the correlation between features. Finally, a compensation strategy is formulated based on the actual power data of historically similar days. This multi-level compensation mechanism effectively solves the problem of power forecast deviation under transitional weather, and improves the forecast accuracy by more than 15% during periods of drastic weather changes.

[0100] By constructing a multi-model fusion architecture of random forest, extreme gradient boosting and long short-term memory network: the random forest model provides stable basic prediction capabilities; the extreme gradient boosting model enhances the ability to identify abnormal situations; the long short-term memory network supplements the ability to capture time series features; this complementary model fusion scheme not only improves the robustness of the prediction, but also reduces the overfitting risk of a single model, and the prediction error is reduced by about 8%.

[0101] Example 2

[0102] Reference Figure 2 - Figure 6 , which is the second embodiment of the present invention, and this embodiment provides a method for predicting the power generation of renewable energy taking into account transitional meteorological conditions. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0103] A method for predicting renewable energy power generation taking into account transitional meteorological conditions comprises the following steps:

[0104] 1) Based on the weather forecast and historical power as input, a new transition weather recognition model is constructed by combining Gram angle field data conversion and introducing efficient channel attention convolutional neural network for feature extraction;

[0105] 2) Use Gaussian mixture model and Mahalanobis distance to classify transitional weather and match it with similar historical days, and formulate transitional weather compensation strategies;

[0106] 3) Design a power prediction fusion model;

[0107] 4) Based on the power forecast fusion model and combined with the transitional weather compensation strategy, a new energy power generation power forecasting method taking into account transitional meteorological conditions is proposed.

[0108] According to the transitional weather identification model designed in step 1), the expression for the Gram angle field data conversion is:

[0109] Assume that the current input one-dimensional time series is X = {x1, x2, x3...x n}, and normalize according to formula (1) to facilitate the conversion to polar coordinate system.

[0110]

[0111] Among them, x i is the i-th value in the time series, is the normalized value, min(X) is the minimum value in the one-dimensional time series X, and max(X) is the maximum value in the one-dimensional time series X.

[0112] The normalized one-dimensional time series X is recorded as Will Mapped to the polar coordinate system, the amplitude is encoded as the cosine of the angle and the time axis is encoded as the radius:

[0113]

[0114] Among them, φ i for The polar angle after the arc cosine function transformation, r i is the polar radius, and n is the number of time points contained in the one-dimensional time series X.

[0115] Finally, the Gram angle field matrix is ​​obtained by calculating the polar angle sum of the time series points after polar coordinate transformation:

[0116]

[0117] The Gram angular field matrix is ​​used to transform the one-dimensional time series into a two-dimensional image with time information, achieving fast and effective feature extraction.

[0118] A convolutional neural network with efficient channel attention is used for feature recognition. After global average pooling of channels without dimensionality reduction, the efficient channel attention mechanism uses one-dimensional convolution to achieve cross-channel information interaction. The mapping between the number of image channels and the convolution kernel is assumed to be a simple nonlinear exponential relationship:

[0119] C=2 γ*k-b ((4)

[0120] Among them, k is the convolution kernel size, C is the number of image channels, γ is the scale factor, and b is the bias term.

[0121] The size of the convolution kernel is adaptively determined by the following function:

[0122]

[0123] Among them, Then x odd is the odd number closest to x.

[0124] The structure of the efficient channel attention convolutional neural network is introduced as follows:

[0125] (1) Input layer: input the image transformed by the Gram angular field matrix;

[0126] (2) Convolutional layer 1: The convolution kernel size is 3×3, the stride is set to 1, and the padding is 1;

[0127] (3) Efficient channel attention layer 1: Perform global average pooling and one-dimensional convolution to achieve local cross-channel interaction. The size of the one-dimensional convolution kernel is adaptively set according to the channel dimension. The activation function uses sigmoid to map the output of the one-dimensional convolution to the interval (0, 1) to obtain the weight of each channel. The output of convolution layer 1 is feature-weighted to dynamically adjust the channel weight.

[0128] (4) Pooling layer 1: The pooling kernel size is 2×2 and the stride is set to 2;

[0129] (5) Convolutional layer 2: The convolution kernel size is 3×3, the stride is set to 1, and the padding is 1;

[0130] (6) Efficient channel attention layer 2: Perform global average pooling and one-dimensional convolution to achieve local cross-channel interaction. The size of the one-dimensional convolution kernel is adaptively set according to the channel dimension. The activation function uses sigmoid to map the output of the one-dimensional convolution to the interval (0, 1) to obtain the weight of each channel. The output of convolution layer 2 is feature-weighted to dynamically adjust the channel weight.

[0131] (7) Pooling layer 2: The pooling kernel size is 2×2 and the stride is set to 2;

[0132] (8) Fully connected layer 1: flatten the output of pooling layer 2;

[0133] (9) Fully connected layer 2: The output of fully connected layer 1 is used as input, and the output is to determine whether the weather type is transitional weather;

[0134] (10) Output layer: turning weather judgment results.

[0135] According to step 2), a transition weather compensation strategy is formulated. For the determined transition weather date, the weather forecast and historical power data are clustered unsupervisedly using the Gaussian mixture model. Let the observed value at time j be x, and its probability is equal to the weighted sum of the probabilities that the observed value belongs to K Gaussian components, expressed as:

[0136]

[0137] Where x is L-dimensional observation data; ω k is the weight coefficient of each Gaussian component in the mixture model; Θ is the set of all parameters of the Gaussian mixture model, Θ={ω k ,μ k ,C k}; N k (x|μ k ,C k ) is the kth Gaussian component of the Gaussian mixture model, whose mean and covariance matrices are μ k and C k .

[0138] Assume that the dataset X∈R N×M , x1, x2 are M-dimensional variables, belonging to two samples in the data set, then the Mahalanobis distance between the two samples is:

[0139]

[0140] Where Σ is the covariance matrix of X.

[0141] According to the results of clustering, matching is performed based on the Mahalanobis distance in the corresponding weather type to obtain historical similar days. The actual power data of historical similar days is subtracted from the current predicted power data to obtain the power prediction compensation sequence.

[0142] The power prediction fusion model structure designed according to step 3) is:

[0143] (1) Basic prediction model layer: The Random Forest (RF) model, the Extreme Gradient Boosting (XGBoost) model and the Long Short-Term Memory (LSTM) neural network model are used as basic prediction models.

[0144] (2) Model fusion layer: Calculate the prediction error of each basic prediction model on the validation set, then normalize it according to the error size to obtain the weight of each model. Finally, the prediction results of the three models are weighted averaged according to the weights to obtain the final prediction result.

[0145] According to step 4), the new energy power generation prediction method considering the transition meteorological conditions is proposed as follows:

[0146] P t =P m,t +δ*P e,t (9)

[0147] Among them, P t is the predicted value of the renewable energy power generation at the tth point; P m,t is the output value of the power prediction fusion model at the tth point; P e,t is the output value of the turning weather compensation strategy at the tth point; δ is the compensation coefficient.

[0148] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0149] As attached Figure 1 To Attachment Figure 6 As shown, the specific embodiment of the present invention includes the following steps:

[0150] The present invention proposes a method for predicting power generation of renewable energy taking into account transitional meteorological conditions. The Gram angle field is combined to convert a one-dimensional time series into a two-dimensional image with time information. A convolutional neural network integrated with an efficient channel self-attention mechanism is introduced to perform efficient feature extraction, and then a judgment of transitional weather is performed. If it is not transitional weather, the meteorological forecast data and historical power data are directly imported into the trained power prediction fusion model to obtain the final prediction result. If it is transitional weather, the Mahalanobis distance is calculated based on the results of the Gaussian mixture model clustering to obtain historical similar days, and the actual power data of historical similar days is subtracted from the current predicted power data to obtain a power prediction compensation sequence. The output of the power prediction fusion model is then combined with the power prediction compensation sequence to obtain the final prediction result. The architecture diagram of this method is shown in the figure. Figure 1 shown.

[0151] The Gram angular field matrix can convert a one-dimensional time series into two-dimensional image data. It not only retains the time information and dynamic features, but also can use image analysis methods to extract features, making the originally complex one-dimensional time series data easy to process and analyze. The expression for Gram angular field data conversion is: Let the current input one-dimensional time series be X = {x1, x2, x3...x n}, and normalized according to formula (10) to facilitate the conversion to polar coordinate system.

[0152]

[0153] Among them, x i is the i-th value in the time series, is the normalized value, min(X) is the minimum value in the one-dimensional time series X, and max(X) is the maximum value in the one-dimensional time series X.

[0154] The normalized one-dimensional time series X is recorded as Will Mapped to the polar coordinate system, the amplitude is encoded as the cosine of the angle and the time axis is encoded as the radius:

[0155]

[0156] Among them, φ i for The polar angle after the arc cosine function transformation, r i is the polar radius, and n is the number of time points contained in the one-dimensional time series X.

[0157] Finally, the Gram angle field matrix is ​​obtained by calculating the polar angle sum of the time series points after polar coordinate transformation:

[0158]

[0159] The power sequence data is transformed using the Gram angular field matrix, and the resulting schematic diagram is as follows Figure 3 As shown, data volatility is reflected in the complexity of image texture and the increasing difference between adjacent pixel values, as well as the large difference between the brightest pixel and the darkest pixel in the image, which has a large dynamic range.

[0160] After the time series data is transformed into an image, a convolutional neural network can be used to extract image features. The convolutional neural network is mainly composed of a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer extracts the features of the input data through a convolution operation. Multiple convolution kernels slide on the input data and perform local weighted summation to extract different features. The pooling layer is generally placed after the convolutional layer and is mainly used to downsample the feature map to reduce the size and amount of calculation of the feature map while retaining important features. After multiple convolutional layers and pooling layers, a fully connected layer is connected to map the extracted feature representation to the sample's label space, integrate the information in the feature map, and output the final classification result.

[0161] In convolutional neural networks, the feature information of different channels in the feature map contributes differently to the final task. The efficient channel attention mechanism compresses spatial information through global average pooling, uses one-dimensional convolution to learn the weight information of the channel, and combines it with the original input feature map to obtain a feature map of specific channel attention.

[0162] The designed convolutional neural network structure with efficient channel attention is as follows Figure 4 As shown, the structure is:

[0163] (1) Input layer: Input the image transformed by the Gram angular field matrix, with an input size of 32×32×3;

[0164] (2) Convolutional layer 1: The convolution kernel size is 3×3, the stride is set to 1, the padding is 1, the number of output channels is 32, and the output size is 32×32×32;

[0165] (3) Efficient channel attention layer 1: Perform global average pooling and one-dimensional convolution to achieve local cross-channel interaction. The size of the one-dimensional convolution kernel is adaptively set according to the channel dimension. The activation function uses sigmoid to map the output of the one-dimensional convolution to the interval (0, 1) to obtain the weight of each channel. The output of convolution layer 1 is feature-weighted to dynamically adjust the weights of the 32 channels.

[0166] (4) Pooling layer 1: The pooling kernel size is 2×2, the stride is set to 2, and the output size is 16×16×32;

[0167] (5) Convolutional layer 2: The convolution kernel size is 3×3, the stride is set to 1, the padding is 1, the number of output channels is 64, and the output size is 16×16×64;

[0168] (6) Efficient channel attention layer 2: Perform global average pooling and one-dimensional convolution to achieve local cross-channel interaction. The size of the one-dimensional convolution kernel is adaptively set according to the channel dimension. The activation function uses sigmoid to map the output of the one-dimensional convolution to the interval (0, 1) to obtain the weight of each channel. The output of convolution layer 2 is feature-weighted to dynamically adjust the weights of the 64 channels.

[0169] (7) Pooling layer 2: The pooling kernel size is 2×2, the stride is set to 2, and the output size is 8×8×64;

[0170] (8) Fully connected layer 1: Flatten the output of pooling layer 2, so the number of input features is 4096 and the number of output features is set to 512;

[0171] (9) Fully connected layer 2: The input feature number is 512, and the output is whether the weather type is transitional weather;

[0172] (10) Output layer: turning weather judgment results.

[0173] If the weather is judged to be transitional weather, the weather forecast and historical power data are clustered unsupervisedly using the Gaussian mixture model. Let the observed value at time j be x, and its probability is equal to the weighted sum of the probabilities that the observed value belongs to K Gaussian components, expressed as:

[0174]

[0175] Where x is L-dimensional observation data; ωk is the weight coefficient of each Gaussian component in the mixture model; Θ is the set of all parameters of the Gaussian mixture model, Θ={ω k ,μ k ,C k}; N k (x|μ k ,C k ) is the kth Gaussian component of the Gaussian mixture model, whose mean and covariance matrices are μ k and C k .

[0176] Assume that the dataset X∈R N×M , x1, x2 are M-dimensional variables, belonging to two samples in the data set, then the Mahalanobis distance between the two samples is:

[0177]

[0178] Where Σ is the covariance matrix of X.

[0179] According to the clustering results, matching is performed in the corresponding weather type based on the Mahalanobis distance to obtain historical similar days. The actual power data of historical similar days is subtracted from the current predicted power data to obtain the power prediction compensation sequence. The prediction situation before compensation, the situation of historical similar days and the power prediction compensation sequence are shown in the figure below. Figure 5 shown.

[0180] The random forest model, extreme gradient boosting model and long short-term memory neural network model are used as basic prediction models. The random forest model randomly selects samples and features, has strong generalization ability and focuses on reducing model variance, while the extreme gradient boosting model focuses on residuals in each training and focuses on reducing model bias. The two integrated algorithms have simple structures and high computational efficiency, but they are not good at processing the autocorrelation between power data. The long short-term memory neural network is good at paying attention to the time series characteristics of power data and can better follow the changes in power, but with the deepening of the network structure and the increase in the number of LSTM units, the amount of calculation will increase greatly. Different models can establish prediction models from different starting points, so the fusion of multiple unrelated models can effectively reduce the prediction error.

[0181] In the model fusion layer, the prediction error of each basic prediction model on the validation set is calculated, and then normalized according to the error size to obtain the weight of each model. Finally, the prediction results of the three models are weighted averaged according to the weights to obtain the final prediction result.

[0182] Based on the power forecast fusion model and combined with the transition weather compensation strategy, a new energy power forecasting method taking into account transition meteorological conditions is proposed:

[0183] P t =P m,t +δ*P e,t (16)

[0184] Among them, P t is the predicted value of the renewable energy power generation at the tth point; P m,t is the output value of the power prediction fusion model at the tth point; P e,t is the output value of the turning weather compensation strategy at the tth point; δ is the compensation coefficient.

[0185] In order to verify the effectiveness of the method proposed in the present invention, a simulation comparative study was conducted on the prediction effects of the traditional single data-driven XGBoost prediction method and the fusion prediction method taking into account the turning meteorological conditions.

[0186] The parameter settings of each model are shown in the following table:

[0187] Table 1 Parameter settings of long short-term memory neural network model

[0188] Parameter name Parameter Value Number of LSTM cells 300 Number of hidden layers 3 Optimization methods Adam Dropout 0.15 Loss Function MSE Activation Function RELU Iterations 500

[0189] Table 2 Random forest model parameter settings

[0190] Parameter name Parameter Value Number of decision trees 500 Maximum tree depth 7 Maximum number of features 5 Minimum number of split samples 2 Minimum number of samples for leaf nodes 2 Split quality index MSE

[0191] Table 3 XGBoost model parameter settings

[0192]

[0193]

[0194] The Gram angular field matrix conversion effect is as follows Figure 3 As shown in the figure, the one-dimensional power generation time series is converted into a two-dimensional image, and the volatility of the data is reflected by the complexity of the image texture and the difference in light and dark contrast between adjacent pixels. After the time series data is converted into an image, a convolutional neural network can be used to extract image features. The convolutional neural network structure with efficient channel attention is introduced as follows: Figure 4 As shown in the figure, the efficient channel attention mechanism dynamically adjusts the weight of each channel, allowing the network to focus on more critical features, thereby improving recognition performance. Figure 5 This is the power forecast compensation effect diagram for transitional meteorological conditions. Through Gaussian mixture model clustering, we find similar historical days and obtain the power forecast compensation sequence, thereby compensating for the originally poor forecast effect under transitional meteorological conditions. Figure 6 As shown by Figure 6It can be seen that the method proposed in the present invention has better following ability under transition meteorological conditions and higher prediction accuracy than the traditional prediction method.

[0195] Compared with the traditional single data driven method, the prediction method designed by the present invention has good prediction accuracy and generalization ability under transition meteorological conditions.

[0196] Example 3

[0197] This embodiment also provides a computer device, which is applicable to a method for predicting renewable energy power generation power taking into account transition meteorological conditions, and includes a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a forced oscillation detection and positioning method for a distribution network as proposed in the above embodiment.

[0198] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, a forced oscillation detection and positioning method for a distribution network is implemented as proposed in the above embodiment.

[0199] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0200] If the function 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 technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0201] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0202] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0203] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0204] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting renewable energy power generation taking into account transitional meteorological conditions, characterized in that: include, Obtain weather forecast and historical power data, use Gram angle field data conversion and efficient channel attention convolutional neural network for feature extraction, and build a turning weather recognition model; According to the discrimination results of the transitional weather identification model, the Gaussian mixture model and Mahalanobis distance are used to classify the transitional weather and match it with similar historical days, and a transitional weather compensation strategy is formulated; Construct a power prediction fusion model based on random forest model, extreme gradient boosting model and long short-term memory neural network; The output result of the power prediction fusion model is combined with the compensation sequence of the transitional weather compensation strategy to obtain the new energy power generation power prediction result.

2. The method for predicting the power generation of renewable energy taking into account transitional meteorological conditions according to claim 1, characterized in that: The Gram angle field data conversion includes: Normalize the input time series; The normalized time series is mapped into a polar coordinate system, where the amplitude is encoded as the angular cosine and the time axis is encoded as the radius; The Gram angle field matrix is ​​obtained by calculating the polar angle sum of the time series points after polar coordinate transformation.

3. The method for predicting the power generation of renewable energy taking into account transitional meteorological conditions as claimed in claim 2, characterized in that: The efficient channel attention convolutional neural network includes: Input layer, used to input the image transformed by the Gram angular field matrix; At least two convolutional layers, each followed by an efficient channel attention layer and a pooling layer; The fully connected layer is used to flatten the output of the pooling layer and output the turning weather discrimination result.

4. The method for predicting the power generation of renewable energy taking into account transitional meteorological conditions as claimed in claim 3, characterized in that: The efficient channel attention layer includes: Feature compression unit, used to perform global average pooling on input features; Feature interaction unit, used to achieve local cross-channel interaction; The weight generation unit is used to generate the weight coefficient of each channel.

5. The method for predicting the power generation of renewable energy taking into account transitional meteorological conditions as claimed in claim 4, characterized in that: The transitional weather compensation strategy includes: The Gaussian mixture model is used to cluster the weather forecast and historical power data; Match historical similar days in corresponding weather types based on Mahalanobis distance; The actual power data of similar historical days is subtracted from the current predicted power data to obtain the power prediction compensation sequence.

6. The method for predicting the power generation of renewable energy taking into account transitional meteorological conditions as claimed in claim 5, characterized in that: The construction process of the power prediction fusion model includes: The random forest model, extreme gradient boosting model and long short-term memory neural network model were established as basic prediction models respectively; Calculate the prediction error of each basic prediction model on the validation set; The weight of each model is obtained by normalizing the prediction error; The prediction results of the three models are weighted averaged according to the weights to obtain the final prediction result.

7. The method for predicting the power generation of renewable energy taking into account transitional meteorological conditions as claimed in claim 6, characterized in that: The new energy power generation power prediction result is obtained by weighted combination of the output value of the power prediction fusion model and the compensation value of the transitional weather compensation strategy.

8. A new energy power generation prediction system taking into account transition meteorological conditions, based on the new energy power generation prediction method taking into account transition meteorological conditions according to any one of claims 1 to 7, characterized in that: Also includes, The feature extraction module obtains weather forecasts and historical power data, uses Gram angle field data conversion and efficient channel attention convolutional neural network to extract features, and builds a transition weather recognition model; A compensation strategy module, based on the discrimination result of the transitional weather identification model, uses a Gaussian mixture model and Mahalanobis distance to classify transitional weather and match it with similar historical days, and formulates a compensation strategy for transitional weather; Prediction fusion module, which builds a power prediction fusion model based on random forest model, extreme gradient boosting model and long short-term memory neural network; The result output module combines the output result of the power prediction fusion model with the compensation sequence of the transitional weather compensation strategy to obtain the new energy power generation power prediction result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for predicting new energy power generation taking into account transition meteorological conditions as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the power generation of new energy taking into account transition meteorological conditions as described in any one of claims 1 to 7 are implemented.

Images