Improved ultra-short-term new energy power prediction system and method considering prediction error

Abstract

The invention belongs to the technical field of power systems, and discloses an improved ultra-short-term new energy power prediction system considering prediction errors, which comprises the following steps: in a first stage, constructing a new energy power prediction model based on VMD-CNN-BiGRU; in the second stage, subtracting the predicted value obtained in the first stage from the true value to obtain an inherent error sequence of the model, taking the inherent error sequence and the selected short-term influence factor as input of the BiGRU network, establishing a nonlinear relationship between the two, and predicting the error by using the model; and adding the prediction results of the two stages to obtain a final prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power systems, and in particular to an improved ultra-short-term new energy power prediction system and method taking prediction errors into consideration. Background Art

[0002] With the continuous pursuit of renewable energy and the continuous advancement of technology, the proportion of wind and solar systems (including wind power and photovoltaic systems) in the power system is increasing. However, the intermittent and volatile nature of wind and solar resources poses challenges to the stable operation of the power grid. Therefore, the grid dispatch response of centralized wind and solar systems considering power forecasting has become the key to ensuring the safe and efficient operation of the power system.

[0003] Power forecasting is to predict the output of wind and solar systems in the future by analyzing meteorological data and historical operating data. Accurate power forecasting can provide an important reference for grid dispatching, helping dispatchers to adjust the output plans of other power sources in advance, balance the supply and demand relationship of the grid, and reduce the impact of wind and solar output fluctuations on the grid. With the large-scale grid connection of renewable distributed energy and energy storage devices, the output fluctuations of new energy have become more random and nonlinear, making the output forecasting of new energy more difficult.

[0004] Traditional prediction methods include regression analysis, time series method, exponential smoothing, grey theory method, etc. These methods have low prediction accuracy and poor nonlinear fitting performance, which cannot meet the growing requirements for new energy output prediction. The accuracy of new energy power prediction depends not only on the selection of input features, but also on the model learning potential and prediction efficiency. Under the same conditions, statistical models with advanced learning functions can often achieve more accurate predictions. Some references use artificial bee colony algorithms to optimize the support vector machine (SVM) model, which significantly improves the learning efficiency and increases the model training efficiency by 30%. There are also studies that combine digital twins with genetic algorithms to optimize BP neural networks, enhance prediction capabilities, and effectively respond to the challenges of limited data samples in emerging power stations. Although shallow learning models such as extreme learning machines and BP neural networks perform well in specific scenarios, they have limitations in revealing the deep correlation between meteorology and photovoltaic power, and have poor adaptability to environmental fluctuations and algorithm robustness. They may also have problems such as slow training convergence, overfitting, or local optimality. Therefore, in order to improve prediction accuracy and model generalization ability, it is crucial to continuously explore and improve the prediction model architecture and algorithm.

[0005] With the advancement of advanced machine learning technology, deep learning frameworks, including long short-term memory networks (LSTM), gated recurrent units, and other deep network structures, have been actively applied in the field of photovoltaic power forecasting. Deep learning models automatically extract complex features from high-dimensional data through multi-layer architectures, explore the nonlinear relationship between meteorological variables and photovoltaic power generation, thereby improving the ability of the forecasting system, which has obvious advantages over traditional shallow neural network models.

[0006] How to avoid the impact of external factors such as time of day, temperature and humidity, day type, and inherent errors in the prediction process of the prediction model on the ultra-short-term renewable energy power prediction performance is an urgent problem to be solved. Summary of the invention

[0007] The embodiments of the present invention provide an improved ultra-short-term new energy power prediction system and method taking into account prediction errors, so as to avoid the influence of external factors such as time of day, temperature and humidity, day type and inherent errors in the prediction process of the prediction model on the ultra-short-term new energy power prediction performance. In order to have a basic understanding of some aspects of the disclosed embodiments, a simple summary is given below. This summary section is not a general review, nor is it intended to identify key / important components or describe the scope of protection of these embodiments. Its only purpose is to present some concepts in a simple form as a preface to the detailed description that follows.

[0008] According to a first aspect of an embodiment of the present invention, there is provided an improved ultra-short-term new energy power prediction system taking prediction errors into consideration.

[0009] In one embodiment, an improved ultra-short-term new energy power prediction system considering prediction error includes:

[0010] In the first stage, a new energy power prediction model based on VMD-CNN-BiGRU is constructed;

[0011] In the second stage, the predicted value obtained in the first stage is subtracted from the true value to obtain the error sequence inherent in the model, which is used together with the selected short-term influencing factors as the input of the BiGRU network, and a nonlinear relationship between the two is established. The model is used to predict the error.

[0012] The final prediction result is obtained by adding the prediction results of the two stages.

[0013] Optionally, in the first stage, VMD is used to decompose the new energy output data with complex fluctuation uncertainty to reduce the complexity of each mode as an input data set of the new energy power prediction model.

[0014] Optionally, in the first stage, based on the BiGRU model that performs well in time series prediction, the CNN convolutional neural network is used to optimize the multivariate data dimensions, and the optimization results are input into the BiGRU model, so that the BiGRU model can perform ultra-short-term predictions based on the characteristics of new energy power data and weather data.

[0015] Optionally, the specific process of the improved ultra-short-term new energy power prediction system considering prediction error includes:

[0016] Step S11, new energy power prediction stage: firstly, the data is normalized, and then the original data sequence is modally decomposed by VMD into various data vectors to increase the dimension of the input data;

[0017] S12, then use the ant colony search algorithm to adaptively optimize the number of hidden layer neurons and learning rate, window size, batch_size, and training rounds of CNN-BiGRU hyperparameters to obtain the optimal parameter group, and use the RAdam algorithm as the gradient optimization algorithm;

[0018] S13, decomposing the data to be predicted into a training set, a validation set, and a test set, using the CNN-BiGRU new energy power prediction model to perform prediction, and accumulating the prediction values ​​of each component to obtain an initial new energy power prediction result;

[0019] S14, subtract the predicted result from the true value to obtain the prediction error result, merge the prediction error sequence with the original data set to form a total data set, and decompose the total data set into a training set, a validation set, and a test set in a ratio of 8:1:1, and input them into CNN-BiGRU to perform multi-step prediction on the prediction error;

[0020] S15, obtaining a prediction result of the prediction error, adding the prediction result to the initial new energy power prediction result, and obtaining a final new energy power prediction result.

[0021] According to a second aspect of an embodiment of the present invention, an improved ultra-short-term new energy power prediction method taking prediction error into consideration is provided.

[0022] In one embodiment, the method comprises:

[0023] Step S21, new energy power data analysis and feature selection;

[0024] Step S22, performing variational mode decomposition on the new energy power sequence;

[0025] Step S23, performing power prediction based on the CNN-BiGRU new energy power prediction method.

[0026] Optionally, the step S21, new energy power data analysis and feature selection, includes:

[0027] From three aspects, namely day type factors, historical renewable energy power and weather factors, we look for factors that affect the change of renewable energy power and use them together with renewable energy power data as input variables of the prediction model.

[0028] Optionally, in the step S21, the Spearman rank correlation coefficient is used to test the correlation between the new energy power and the weather information, and the influencing factors of the new energy power prediction are selected according to the correlation coefficient value.

[0029] Optionally, the step S22, performing variational mode decomposition on the new energy power sequence, includes:

[0030] Variational modal decomposition is used to decompose the highly volatile renewable energy power sequence into several simple and relatively stable modal components. Through variational modal decomposition, the original signal is effectively decomposed into several modal components.

[0031] Optionally, the step S22 specifically includes:

[0032] First, the new energy power data is sampled to obtain a new energy power data sequence f(t);

[0033] Then, variational mode decomposition is used to decompose f(t) into mode functions u with different frequency characteristics. k , without considering the decomposition loss, the sum of the modal components is the initial input f(t), and the constrained variational problem constructed from this is:

[0034]

[0035] Where K is the number of target modes decomposed by VMD; {u k}、{ω k} are the mode sets and center frequency sets respectively; δ(t) is the Dirac distribution function.

[0036] Optionally, the step S22 further includes: introducing a Lagrangian multiplication operator λ(t) and a quadratic balance parameter α, transforming the constrained problem in equation (1) into an unconstrained problem, and obtaining the augmented Lagrangian equation as follows:

[0037]

[0038] The quadratic balance parameter α ensures accurate signal reconstruction under the influence of Gaussian noise; the Lagrange multiplication operator λ(t) ensures the strictness of the constraints.

[0039] Optionally, the step S22 further includes:

[0040] The unconstrained variational problem iteratively updates each modal component and its central frequency through the cross multiplier algorithm, and finally obtains the saddle point of the unconstrained model, which is also the optimal solution of the variational problem.

[0041] Optionally, {u k}、{ω k The iterative update formula of} is as follows:

[0042]

[0043] In the formula, They are the Wiener filter and frequency center corresponding to each modal component at the n+1th iteration.

[0044] Optionally, {u k}、{ω k The iterations of} are continued until the following convergence conditions are met:

[0045]

[0046] Where ε is the convergence accuracy, which is a constant greater than 0.

[0047] Optionally, the step S23, performing power prediction based on the CNN-BiGRU new energy power prediction method, includes:

[0048] The CNN input layer obtains several modal components collected by the input layer and extracts feature vectors through CNN.

[0049] Optionally, the step S23 further includes performing bidirectional network training using BiGRU to capture the bidirectional dependency relationship between historical new energy power data and future new energy power data.

[0050] According to a third aspect of an embodiment of the present invention, a computer device is provided.

[0051] In some embodiments, the computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0052] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0053] It can make ultra-short-term forecasts quickly and accurately based on the characteristics of new energy power data and weather data.

[0054] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0056] Figure 1 is a specific flow chart of an improved ultra-short-term new energy power prediction system considering prediction errors according to an exemplary embodiment;

[0057] Figure 2 is a flow chart of an improved ultra-short-term new energy power prediction method considering prediction error according to an exemplary embodiment;

[0058] Figure 3 is a schematic structural diagram of a GRU module according to an exemplary embodiment;

[0059] Figure 4 is a schematic diagram of a bidirectional cyclic structure of a BiGRU network at times t-1, t, and t+1 according to an exemplary embodiment;

[0060] Figure 5 It is a schematic diagram showing the structure of a computer device according to an exemplary embodiment. DETAILED DESCRIPTION

[0061] The following description and accompanying drawings fully illustrate the specific embodiments of this article so that those skilled in the art can practice them. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. The scope of the embodiments of this article includes the entire scope of the claims, as well as all available equivalents of the claims. Herein, the terms "first", "second", etc. are only used to distinguish one element from another, without requiring or implying any actual relationship or order between these elements. In fact, the first element can also be called the second element, and vice versa. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the structure, device or equipment including a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also include elements inherent to such structure, device or equipment. In the absence of more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the structure, device or equipment including the elements. Each embodiment is described in a progressive manner herein, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the embodiments can be referred to each other.

[0062] The terms "longitudinal", "lateral", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc. in this document indicate the orientation or position relationship based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing this document and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In the description of this document, unless otherwise specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a mechanical connection or an electrical connection, it can also be the internal communication of two elements, it can be a direct connection, or it can be an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0063] As used herein, the term "plurality" means two or more than two, unless otherwise specified.

[0064] In this document, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B means: A or B.

[0065] In this article, the term "and / or" is a description of the association relationship between objects, indicating that three relationships may exist. For example, A and / or B means: A or B, or, A and B.

[0066] It should be understood that, although the various steps in the flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the figure may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0067] Each module in the device or system of the present application can be implemented in whole or in part by software, hardware, or a combination thereof. The above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above modules.

[0068] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0069] The present invention proposes an improved ultra-short-term new energy power prediction system considering prediction errors. In the first stage, a new energy power prediction model based on VMD-CNN-BiGRU is constructed, and VMD is used to decompose new energy output data with complex fluctuation uncertainty to reduce the complexity of each mode as the input data set of the new energy output prediction model. Based on the BiGRU model that performs well in time series prediction, the CNN convolutional neural network is used to optimize the multivariate data dimension, and the optimization result is input into the BiGRU model, so that the BiGRU model can quickly and accurately perform ultra-short-term prediction according to the characteristics of new energy power data and weather data. In the second stage, the predicted value obtained in the first stage is subtracted from the true value to obtain the error sequence inherent in the model, and the error sequence is used together with the selected short-term influencing factors as the input of the BiGRU network, and a nonlinear relationship between the two is established. The model is used to realize the prediction of the error, and finally the prediction results of the two stages are added together to obtain the final prediction result.

[0070] like Figure 1 As shown, the specific process of the improved ultra-short-term new energy power prediction system considering prediction error includes:

[0071] Step S11, i.e., the new energy power prediction stage: first, the data is normalized, and then the original data sequence is modally decomposed using VMD to decompose it into various data vectors, thereby increasing the dimension of the input data;

[0072] S12, then use the ant colony search algorithm to adaptively optimize the number of hidden layer neurons and learning rate, window size, batch_size, and training rounds of CNN-BiGRU hyperparameters to obtain the optimal parameter group, and use the RAdam algorithm as the gradient optimization algorithm. The RAdam algorithm has stronger stability than the Adam algorithm and the traditional gradient descent algorithm, and provides a dynamic heuristic method to provide automated variance attenuation, thereby eliminating the need for manual tuning during training;

[0073] S13, decomposing the data to be predicted into a training set, a validation set, and a test set, using the CNN-BiGRU new energy power prediction model to perform prediction, and accumulating the prediction values ​​of each component to obtain an initial new energy power prediction result;

[0074] S14, subtract the predicted result from the true value to obtain the prediction error result, merge the prediction error sequence with the original data set to form a total data set, and decompose the total data set into a training set, a validation set, and a test set in a ratio of 8:1:1, and input them into CNN-BiGRU to perform multi-step prediction on the prediction error;

[0075] S15, obtaining a prediction result of the prediction error, adding the prediction result to the initial new energy power prediction result, and obtaining a final new energy power prediction result.

[0076] An improved ultra-short-term new energy power prediction method taking prediction error into consideration according to the present invention is described in detail below with reference to the accompanying drawings.

[0077] Figure 2 An embodiment of an improved ultra-short-term new energy power prediction method taking prediction error into consideration according to the present invention is shown.

[0078] In this optional embodiment, the improved ultra-short-term new energy power prediction method considering prediction error includes:

[0079] Step S21, new energy power data analysis and feature selection.

[0080] The prediction of new energy power is related to many factors. The present invention mainly searches for factors affecting the change of new energy power from three aspects: day type factors, historical new energy power and weather factors (temperature, precipitation, wind direction, wind speed, humidity, dew point, air pressure, solar zenith angle, horizontal total radiation, scattered radiation, direct radiation and surface albedo), and uses them together with new energy power data as input variables of the prediction model, thereby improving the accuracy of the prediction results. Appropriate input variables are directly related to the prediction performance of the model and are an important part of the new energy power prediction model. The input variables should contain enough influencing factors. Too many input variables will increase the training burden and affect the training efficiency. Since the new energy power is affected by weather factors and has strong randomness and instability, it is necessary to conduct a correlation analysis between the new energy power and weather factors to effectively select the input variables of the new energy power prediction model.

[0081] Optionally, in the above step S21, the Spearman rank correlation coefficient is used to test the correlation between the new energy power and the weather information, and the influencing factors of the new energy power prediction are selected according to the correlation coefficient value.

[0082] Step S22, performing variational mode decomposition on the new energy power sequence.

[0083] In order to avoid overfitting during the training of the new energy power prediction model and reduce the complexity of the model input, the present invention uses VMD to decompose the highly volatile new energy power sequence into several simple and relatively stable modal components. Through variational mode decomposition (VMD), the original signal is effectively decomposed into several modal components. In this process, each modal component is regarded as an amplitude-frequency modulated signal.

[0084] First, the new energy power data is sampled to obtain a new energy power data sequence f(t), and then VMD is used to decompose f(t) into modal functions u with different frequency characteristics. k , without considering the decomposition loss, the sum of the modal components is the initial input f(t), and the constrained variational problem constructed from this is:

[0085]

[0086] Where K is the number of target modes decomposed by VMD; {u k}、{ω k} are the mode sets and center frequency sets respectively; δ(t) is the Dirac distribution function.

[0087] By introducing the Lagrangian multiplication operator λ(t) and the quadratic balance parameter α, the constrained problem in equation (1) is transformed into an unconstrained problem, and the augmented Lagrangian equation is obtained as follows:

[0088]

[0089] The quadratic balance parameter α can ensure that the signal can be accurately reconstructed under the influence of Gaussian noise; the Lagrange multiplication operator λ(t) can ensure the strictness of the constraints.

[0090] Unconstrained variational problems can be solved by iteratively updating each modal component and its central frequency through the cross multiplier algorithm, and finally obtaining the saddle point of the unconstrained model, which is also the optimal solution to the variational problem. k}、{ω k The iterative update formula of} is as follows:

[0091]

[0092] In the formula, are the Wiener filters and frequency centers corresponding to each modal component at the n+1th iteration. The iteration is performed until the following convergence conditions are met:

[0093]

[0094] Where ε is the convergence accuracy, which is a constant greater than 0.

[0095] Step S23, performing power prediction based on the CNN-BiGRU new energy power prediction method.

[0096] The CNN input layer obtains several modal components collected by the input layer, extracts feature vectors through CNN, and uses BiGRU for bidirectional network training, which effectively captures the bidirectional dependency between historical renewable energy power data and future renewable energy power data, and improves the utilization efficiency of data features and the accuracy and robustness of model prediction.

[0097] The GRU neural network is a variation of the recurrent neural network. The GRU unit includes a forget gate, an input gate, and an output gate. Assuming that it represents all the inputs of a long short-term memory network, the new energy power data and weather data are first decomposed by VMD into K modal components, and these K modal components are respectively input into the CNN-BiGRU network for training, then the value corresponding to the modal component at time t is represented. The structure of the GRU module corresponding to the time t is as follows: Figure 3 shown.

[0098] Recurrent Neural Network (RNN) is a model used to process time series data. The parameters of GRU neural network are reduced by 1 / 3 compared with LSTM, which is not easy to produce overfitting. In addition, GRU neural network is faster in terms of convergence time and number of required iterations. GRU can make the structure of the model simpler while maintaining the effect of LSTM. The forward propagation formula of CNN-BiGRU neural network is as follows:

[0099] h k =RELU(W k *X+b k ) (6)

[0100] After the input data passes through the convolution layer, the output matrix enters the recurrent layer and the recurrent skip layer. The recurrent layer and the recurrent skip layer use the RELU function as the activation function of the hidden layer. At time t, the hidden state of the recurrent unit in the recurrent layer is:

[0101] r t =σ(x t W xr +h t-1 W hr +b r ) (7)

[0102] u t =σ(x t W xu +h t-1 W hu +b u ) (8)

[0103] c t =RELU(x t W xc +rt ⊙(h t-1 W hc )+b c ) (9)

[0104] c t =RELU(x t W xc +r t ⊙(h t-1 W hc )+b c ) (10)

[0105] h t =(1-u t )⊙h t-p +u t ⊙c t (11)

[0106] Where ⊙ represents the element-wise multiplication, σ is the sigmoid function, and x t is the data input at time t, r t 、u t 、c t are the reset gate, update gate, and candidate hidden state values ​​at time t, respectively. W and b are the weights and biases of each module, respectively.

[0107] The advantages of choosing the BiGRU model are mainly reflected in its ability to process time series data and its computational efficiency. First, BiGRU can simultaneously consider the past and future information of time series data, and use the context information to better capture the characteristics of time series, which makes the model perform well when processing data with complex dependencies.

[0108] Figure 4 It is the bidirectional loop structure of the BiGRU network at t-1, t, and t+1, where x is the data input of the model, h is the hidden layer state of the network, and y is the data output. BiGRU can perform recursive loops in both positive and negative time flows, so there are hidden layer states in two directions, and there is no interaction between the hidden layers in these two directions. It can be split into two independent GRU networks with opposite data time flows.

[0109] set up is the hidden state of the forward GRU network, then

[0110]

[0111] set up is the hidden state of the reverse GRU network, then

[0112]

[0113] In the formula, each decomposed mode is input into two GRU networks in positive order and negative order respectively. and h← t The combination of these two parts of the hidden layer state output value is the output of the BiGRU network. Since there is an error between the predicted value and the actual power, in order to further improve the prediction accuracy, it is necessary to compensate the error of the initial new energy power prediction value in the first stage. Using the new energy power prediction results obtained by VMD-CNN-BiGRU, the prediction error sequence at each time point is extracted, and then together with the factors that cause the fluctuation and nonlinearity of the new energy power (temperature, humidity, day type factors) to form an error training set as the input of the error compensation model, and a multi-step CNN-BiGRU prediction error model is established.

[0114] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, 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, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in the above method embodiment are implemented.

[0115] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0116] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiment when executing the computer program.

[0117] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiment are implemented.

[0118] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0119] The present invention is not limited to the structures which have been described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. An improved ultra-short-term new energy power prediction system considering prediction error, characterized in that: include: In the first stage, a new energy power prediction model based on VMD-CNN-BiGRU is constructed; In the second stage, the predicted value obtained in the first stage is subtracted from the true value to obtain the error sequence inherent in the model, which is used together with the selected short-term influencing factors as the input of the BiGRU network, and a nonlinear relationship between the two is established. The model is used to predict the error. The final prediction result is obtained by adding the prediction results of the two stages.

2. The improved ultra-short-term new energy power prediction system considering prediction error as claimed in claim 1, characterized in that: In the first stage, VMD is used to decompose the new energy output data with complex fluctuation uncertainty, reduce the complexity of each mode, and use it as the input data set of the new energy power prediction model.

3. The improved ultra-short-term new energy power prediction system considering prediction error as described in claim 2, characterized in that: In the first stage, based on the BiGRU model that performs well in time series prediction, the CNN convolutional neural network is used to optimize the multivariate data dimensions, and the optimization results are input into the BiGRU model, so that the BiGRU model can perform ultra-short-term predictions based on the characteristics of new energy power data and weather data.

4. The improved ultra-short-term new energy power prediction system considering prediction error as claimed in claim 1, characterized in that: The specific process of the improved ultra-short-term new energy power prediction system considering prediction error includes: Step S11, new energy power prediction stage: firstly, the data is normalized, and then the original data sequence is modally decomposed by VMD into various data vectors to increase the dimension of the input data; S12, then use the ant colony search algorithm to adaptively optimize the number of hidden layer neurons and learning rate, window size, batch_size, and training rounds of CNN-BiGRU hyperparameters to obtain the optimal parameter group, and use the RAdam algorithm as the gradient optimization algorithm; S13, decomposing the data to be predicted into a training set, a validation set, and a test set, using the CNN-BiGRU new energy power prediction model to perform prediction, and accumulating the prediction values ​​of each component to obtain an initial new energy power prediction result; S14, subtract the predicted result from the true value to obtain the prediction error result, merge the prediction error sequence with the original data set to form a total data set, and decompose the total data set into a training set, a validation set, and a test set in a ratio of 8:1:1, and input them into CNN-BiGRU to perform multi-step prediction on the prediction error; S15, obtaining a prediction result of the prediction error, adding the prediction result to the initial new energy power prediction result, and obtaining a final new energy power prediction result.

5. An improved ultra-short-term new energy power prediction method considering prediction error, characterized in that: include: Step S21, new energy power data analysis and feature selection; Step S22, performing variational mode decomposition on the new energy power sequence; Step S23, performing power prediction based on the CNN-BiGRU new energy power prediction method.

6. The improved ultra-short-term new energy power prediction method considering prediction error as claimed in claim 5, characterized in that: The step S21, new energy power data analysis and feature selection, includes: From three aspects, namely day type factors, historical renewable energy power and weather factors, we look for factors that affect the change of renewable energy power and use them together with renewable energy power data as input variables of the prediction model.

7. The improved ultra-short-term new energy power prediction method considering prediction error as claimed in claim 6, characterized in that: In the step S21, the Spearman rank correlation coefficient is used to test the correlation between the new energy power and the weather information, and the influencing factors of the new energy power prediction are selected according to the correlation coefficient value.

8. The improved ultra-short-term new energy power prediction method considering prediction error as claimed in claim 5, characterized in that: The step S22, performing variational mode decomposition on the new energy power sequence, comprises: Variational modal decomposition is used to decompose the highly volatile renewable energy power sequence into several simple and relatively stable modal components. Through variational modal decomposition, the original signal is effectively decomposed into several modal components.

9. The improved ultra-short-term new energy power prediction method considering prediction error as claimed in claim 8, characterized in that: The step S22 specifically includes: First, the new energy power data is sampled to obtain a new energy power data sequence f(t); Then, variational mode decomposition is used to decompose f(t) into mode functions u with different frequency characteristics. k , without considering the decomposition loss, the sum of the modal components is the initial input f(t), and the constrained variational problem constructed from this is: Where K is the number of target modes decomposed by VMD; {u k }、{ω k } are the mode sets and center frequency sets respectively; δ(t) is the Dirac distribution function.

10. The improved ultra-short-term new energy power prediction method considering prediction error according to claim 9, characterized in that: The step S22 further includes: introducing the Lagrangian multiplication operator λ(t) and the quadratic balance parameter α, transforming the constrained problem in equation (1) into an unconstrained problem, and obtaining the augmented Lagrangian equation as follows: The quadratic balance parameter α ensures accurate signal reconstruction under the influence of Gaussian noise; the Lagrange multiplication operator λ(t) ensures the strictness of the constraints.

11. The improved ultra-short-term new energy power prediction method considering prediction error according to claim 10, characterized in that: The step S22 further includes: The unconstrained variational problem iteratively updates each modal component and its central frequency through the cross multiplier algorithm, and finally obtains the saddle point of the unconstrained model, which is also the optimal solution of the variational problem.

12. The improved ultra-short-term new energy power prediction method considering prediction error according to claim 11, characterized in that: {u k }、{ω k The iterative update formula of} is as follows: In the formula, They are the Wiener filter and frequency center corresponding to each modal component at the n+1th iteration.

13. The improved ultra-short-term new energy power prediction method considering prediction error according to claim 12, characterized in that: {u k }、{ω k The iterations of} are continued until the following convergence conditions are met: Where ε is the convergence accuracy, which is a constant greater than 0.

14. The improved ultra-short-term new energy power prediction method considering prediction error according to claim 5, characterized in that: The step S23, performing power prediction based on the CNN-BiGRU new energy power prediction method, includes: The CNN input layer obtains several modal components collected by the input layer and extracts feature vectors through CNN.

15. The improved ultra-short-term new energy power prediction method considering prediction error according to claim 14, characterized in that: The step S23 also includes using BiGRU to perform bidirectional network training to capture the bidirectional dependency relationship between historical new energy power data and future new energy power data.

16. 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 according to any one of claims 5 to 15 are implemented.

Images