New energy power prediction method and device based on time sequence decomposition

Abstract

The invention provides a new energy power prediction method and device based on time sequence decomposition. The method comprises the steps of obtaining an original time sequence of target new energy; performing time domain decomposition on the original time sequence to obtain a trend term, a seasonal term and a residual term, and performing frequency domain decomposition on the residual term to obtain modal features; performing feature screening based on the trend item and the season item to obtain target feature data; and applying the target feature data, the modal features and a preset power prediction model to determine a power prediction result of the target new energy, the preset power prediction model being obtained by pre-training a bidirectional long-short-term memory network model based on batch historical original time sequences and respective corresponding actual power thereof. According to the invention, the reliability of wind and light power prediction can be improved on the basis of ensuring the efficiency of wind and light power prediction.

Description

Technical Field

[0001] The present application relates to the technical field of power prediction, and in particular to a method and device for predicting new energy power based on time series decomposition. Background Art

[0002] Renewable energy sources such as wind power and photovoltaic power generation (abbreviated as wind and solar) are clean and pollution-free and are considered to be important components of future power systems. However, wind power and photovoltaic power generation have significant intermittent and random characteristics, which makes their power in the power system difficult to predict. Therefore, accurate prediction of wind power and photovoltaic power generation is very important for power system scheduling and operation.

[0003] At present, the power forecasting method for wind and solar power is usually to use machine learning models as power forecasting models for power forecasting. For example, ARIMA and SARIMA models have good performance when processing linear time series, but they perform poorly when dealing with nonlinear and non-stationary data. In addition, the power forecasting of wind and solar power is affected by many factors, such as meteorological conditions, environmental changes, and equipment status, so the data has high complexity and nonlinear characteristics. At present, feature selection is ignored when processing and forecasting wind and solar power, resulting in poor performance of the power forecasting model and affecting the reliability of wind and solar power forecasting. Summary of the invention

[0004] In response to at least one problem in the prior art, the present application proposes a new energy power prediction method and device based on time series decomposition, which can improve the reliability of wind and solar power prediction while ensuring the efficiency of wind and solar power prediction.

[0005] In order to solve the above technical problems, this application provides the following technical solutions:

[0006] In a first aspect, the present application provides a new energy power prediction method based on time series decomposition, comprising:

[0007] Obtain the original time series of the target new energy;

[0008] Decomposing the original time series in the time domain to obtain trend terms, seasonal terms and residual terms, and decomposing the residual terms in the frequency domain to obtain modal features;

[0009] Perform feature screening based on the trend item and the season item to obtain target feature data;

[0010] The target feature data, modal features and a preset power prediction model are applied to determine the power prediction result of the target new energy. The preset power prediction model is obtained by pre-training a bidirectional long short-term memory network model based on batch historical original time series and their respective corresponding actual powers.

[0011] In one embodiment, the new energy power prediction method based on time series decomposition further includes:

[0012] Collect batch historical raw time series and their corresponding actual power;

[0013] Perform time domain decomposition on each historical original time series to obtain the historical trend term, historical seasonal term and historical residual term corresponding to the historical original time series, and perform frequency domain decomposition on the historical residual term corresponding to each historical original time series to obtain the historical modal characteristics of the historical original time series;

[0014] Perform feature screening based on the historical trend item and the historical season item corresponding to each historical original time series to obtain historical feature data of the historical original time series;

[0015] The bidirectional long short-term memory network model is trained using the historical feature data, historical modal features and actual power corresponding to each historical original time series to obtain the preset power prediction model.

[0016] In one embodiment, the performing frequency domain decomposition on the residual term to obtain the modal feature includes:

[0017] The variational mode decomposition method is applied to decompose the residual term in the frequency domain to obtain the mode characteristics.

[0018] In one embodiment, the feature screening based on the trend item and the season item to obtain target feature data includes:

[0019] The maximum mutual information coefficient is applied to perform feature screening on the trend item and the seasonal item to obtain target feature data.

[0020] In one embodiment, applying the target characteristic data, the modal characteristics and a preset power prediction model to determine the power prediction result of the target new energy source includes:

[0021] Applying the target characteristic data and a preset first sub-power prediction model to determine a first intermediate power prediction value;

[0022] Applying the modal characteristics and a preset second sub-power prediction model to determine a second intermediate power prediction value;

[0023] Determine the sum of the first intermediate power prediction value and the second intermediate power prediction value as the power prediction result of the target new energy;

[0024] The preset power prediction model includes: the preset first sub-power prediction model and the preset second sub-power prediction model.

[0025] In one embodiment, the application of the historical feature data, historical modal features and actual power corresponding to each historical original time series to train the bidirectional long short-term memory network model to obtain the preset power prediction model includes:

[0026] Input the historical feature data into a first bidirectional long short-term memory network model to obtain the output result of the first bidirectional long short-term memory network model, input the historical modal features into a second bidirectional long short-term memory network model to obtain the output result of the second bidirectional long short-term memory network model; use the sum of the output results of the first bidirectional long short-term memory network model and the second bidirectional long short-term memory network model and the corresponding actual power to train and obtain the preset power prediction model;

[0027] Among them, the preset first sub-power prediction model is a trained first bidirectional long short-term memory network model, and the second sub-power prediction model is a trained second bidirectional long short-term memory network model.

[0028] In a second aspect, the present application provides a new energy power prediction device based on time series decomposition, comprising:

[0029] An acquisition module is used to obtain the original time series of the target new energy;

[0030] A first decomposition module is used to perform time domain decomposition on the original time series to obtain trend terms, seasonal terms and residual terms, and perform frequency domain decomposition on the residual terms to obtain modal features;

[0031] A first screening module, used for performing feature screening based on the trend item and the season item to obtain target feature data;

[0032] A prediction module is used to apply the target feature data, modal features and a preset power prediction model to determine the power prediction result of the target new energy. The preset power prediction model is obtained by pre-training a bidirectional long short-term memory network model based on batch historical original time series and their respective corresponding actual powers.

[0033] In one embodiment, the new energy power prediction device based on time series decomposition further includes:

[0034] The acquisition module is used to collect batch historical original time series and their corresponding actual power;

[0035] The second decomposition module is used to perform time domain decomposition on each historical original time series to obtain the historical trend term, historical seasonal term and historical residual term corresponding to the historical original time series, and perform frequency domain decomposition on the historical residual term corresponding to each historical original time series to obtain the historical modal features of the historical original time series;

[0036] A second screening module is used to perform feature screening based on the historical trend item and the historical season item corresponding to each historical original time series to obtain historical feature data of the historical original time series;

[0037] The training module is used to train the bidirectional long short-term memory network model by applying the historical feature data, historical modal features and actual power corresponding to each historical original time series to obtain the preset power prediction model.

[0038] In one embodiment, the first decomposition module comprises:

[0039] The frequency domain decomposition unit is used to apply the variational mode decomposition method to perform frequency domain decomposition on the residual term to obtain the mode feature.

[0040] In one embodiment, the first screening module includes:

[0041] The screening unit is used to apply the maximum mutual information coefficient to perform feature screening on the trend item and the seasonal item to obtain target feature data.

[0042] In one embodiment, the prediction module includes:

[0043] A first determining unit, configured to determine a first intermediate power prediction value by applying the target characteristic data and a preset first sub-power prediction model;

[0044] A second determining unit, configured to determine a second intermediate power prediction value by applying the modal feature and a preset second sub-power prediction model;

[0045] Determine the sum of the first intermediate power prediction value and the second intermediate power prediction value as the power prediction result of the target new energy;

[0046] The prediction unit is used for the preset power prediction model including: the preset first sub-power prediction model and the preset second sub-power prediction model.

[0047] In one embodiment, the training module includes:

[0048] An input unit, used to input the historical feature data into a first bidirectional long short-term memory network model to obtain an output result of the first bidirectional long short-term memory network model, and input the historical modal features into a second bidirectional long short-term memory network model to obtain an output result of the second bidirectional long short-term memory network model;

[0049] A training unit, configured to train the preset power prediction model using the sum of the output results of the first bidirectional long short-term memory network model and the second bidirectional long short-term memory network model and the corresponding actual power;

[0050] Among them, the preset first sub-power prediction model is a trained first bidirectional long short-term memory network model, and the second sub-power prediction model is a trained second bidirectional long short-term memory network model.

[0051] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the new energy power prediction method based on time series decomposition when executing the program.

[0052] In a fourth aspect, the present application provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the new energy power prediction method based on time series decomposition.

[0053] It can be seen from the above technical solution that the present application provides a method and device for predicting power of new energy based on time series decomposition. Among them, the method includes: obtaining the original time series of the target new energy; performing time domain decomposition on the original time series to obtain trend terms, seasonal terms and residual terms, performing frequency domain decomposition on the residual terms to obtain modal features; performing feature screening based on the trend terms and seasonal terms to obtain target feature data; applying the target feature data, modal features and a preset power prediction model to determine the power prediction result of the target new energy, the preset power prediction model is based on the batch historical original time series and their respective corresponding actual power pairs. The bidirectional long short-term memory network model is pre-trained, which can improve the reliability of wind and solar power prediction on the basis of ensuring the efficiency of wind and solar power prediction, thereby improving the stability and economy of the power system, and can provide a prediction model that effectively processes complex time series data, and can also effectively reduce the demand for spare capacity and peak load regulation, and realize efficient and accurate wind and solar power prediction, which is of great significance for the optimized operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 It is a first flow chart of a new energy power prediction method based on time series decomposition in an embodiment of the present application;

[0056] Figure 2 It is a second flow chart of the new energy power prediction method based on time series decomposition in the embodiment of the present application;

[0057] Figure 3 It is a schematic diagram comparing photovoltaic power prediction results of multiple cases in the application example of this application;

[0058] Figure 4 It is a schematic diagram comparing wind power prediction results of multiple cases in the application example of this application;

[0059] Figure 5 It is a first structural schematic diagram of a new energy power prediction device based on time series decomposition in an embodiment of the present application;

[0060] Figure 6 It is a second structural schematic diagram of a new energy power prediction device based on time series decomposition in an embodiment of the present application;

[0061] Figure 7 It is a schematic block diagram of the system structure of the electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0063] In order to solve the problems existing in the above-mentioned prior art, the embodiment of the present application provides a new energy power prediction method and device based on time series decomposition, which can be based on the wind-light short-term power BiLSTM prediction framework of time-frequency decomposition. The framework uses the maximum mutual information coefficient MIC for feature screening by integrating time domain decomposition and frequency domain decomposition technology, and uses the BiLSTM model for prediction, thereby achieving significant improvements in multi-scale feature capture, effectiveness of feature selection, and efficiency and accuracy of long-sequence prediction, and can more accurately predict the short-term power of wind-light new energy.

[0064] (1) The embodiment of the present application can construct a wind-solar forecast data processing framework that integrates time domain decomposition and frequency domain decomposition. The framework uses the STL (Seasonal and Trend decomposition using Loess) method for time domain decomposition to decompose the original time series into trend terms, seasonal terms, and residual terms; then the residual terms are secondary decomposed using variational mode decomposition (VMD) to extract modal functions of different frequencies. This multi-scale data processing framework excels in capturing multi-scale features in the data and can improve the accuracy and stability of the prediction model. That is, the embodiment of the present application can combine time domain decomposition or frequency domain decomposition to fully utilize the advantages of both. By integrating the STL and variational mode decomposition (VMD) methods, the characteristics of wind power and photovoltaic power data are fully captured, thereby improving the accuracy and stability of the prediction model.

[0065] (2) The embodiment of the present application proposes a feature screening method based on the maximum mutual information coefficient (MIC), which aims to screen out feature variables that are highly correlated with wind power and photovoltaic power generation. The MIC method can effectively capture the linear and nonlinear relationships between variables, and has higher robustness and adaptability than the traditional correlation coefficient method. By using MIC for feature screening, it is ensured that the data input into the prediction model has a high correlation, thereby significantly improving the prediction performance and reliability of the model. That is, considering that wind power and photovoltaic power generation are affected by many factors, such as meteorological conditions, environmental changes, and equipment status, the data has a high degree of complexity and nonlinear characteristics. The embodiment of the present application introduces a feature selection step, based on the feature selection method of the maximum mutual information coefficient (MIC), effectively identifying and quantifying the linear and nonlinear relationships between variables, thereby screening out features that are highly correlated with wind power and photovoltaic power generation, thereby improving the performance and reliability of the prediction model. It is possible to extract the most representative features from a large amount of raw data, reduce redundant information, and improve the performance and computational efficiency of the prediction model.

[0066] (3) The present application uses the BiLSTM model for short-term power forecasting of wind power and photovoltaic power generation. The BiLSTM model is a bidirectional neural network model designed for long-term series prediction. The core of the BiLSTM model is the introduction of a gating mechanism, which can better capture the temporal dependencies in the sequence and greatly reduce the computational complexity. Compared with the traditional sequence-to-sequence model, the BiLSTM model shows higher efficiency and accuracy when processing long sequence data, and is particularly suitable for wind power and photovoltaic power generation prediction tasks with complex temporal characteristics. That is, considering that the traditional prediction model performs poorly when dealing with nonlinear and non-stationary data, has high computational complexity when processing long sequence data, and has a long training time; the present application embodiment considers the application of an improved recurrent neural network model BiLSTM, which can significantly improve the prediction effect of the time series by mining the characteristics of bidirectional time series data to obtain temporal dependencies, realize short-term output prediction of wind power and photovoltaic power generation, and improve the efficiency and accuracy of the prediction by optimizing the model structure and algorithm.

[0067] The details are described through the following embodiments.

[0068] In order to improve the reliability of wind and solar power prediction on the basis of ensuring the efficiency of wind and solar power prediction, this embodiment provides a new energy power prediction method based on time series decomposition, in which the execution subject is a new energy power prediction device based on time series decomposition, and the new energy power prediction device based on time series decomposition includes but is not limited to a server, such as Figure 1 As shown, the method specifically includes the following contents:

[0069] Step 100: Obtain the original time series of the target new energy.

[0070] Specifically, the target new energy source may be a wind farm or a photovoltaic power station. For example, the original time series of a photovoltaic power station may be photovoltaic power data collected at preset time intervals; the original time series of a wind farm may be wind power data collected at preset time intervals; the preset time interval may be set according to actual conditions, and this application does not limit this.

[0071] Step 200: Decompose the original time series in the time domain to obtain trend terms, seasonal terms and residual terms, and decompose the residual terms in the frequency domain to obtain modal features.

[0072] In order to improve the accuracy of capturing long-term trends and seasonal fluctuations, adapt to seasonal changes of different frequencies and amplitudes, especially wind-light time series data with complex seasonality and nonlinear trends, the time series decomposition method STL can be applied to decompose the original time series in the time domain to obtain trend terms, seasonal terms and residual terms. By performing frequency domain decomposition on the residual terms to obtain modal characteristics, the problem of large randomness of residual term fluctuations and high difficulty in prediction can be avoided.

[0073] Step 300: Perform feature screening based on the trend item and the season item to obtain target feature data.

[0074] Specifically, the power generation can be used as the target variable, and the features with high correlation with the power generation can be selected as the target feature data; if the target new energy source is a wind farm, the target feature data can be wind speed, temperature and other features with high correlation with the power generation. If the target new energy source is a photovoltaic power station, the target feature data can be radiation, sun angle and other features with high correlation with the power generation.

[0075] Step 400: Apply the target feature data, modal features and a preset power prediction model to determine the power prediction result of the target new energy. The preset power prediction model is pre-trained based on a bidirectional long short-term memory network model based on batch historical original time series and their respective corresponding actual powers.

[0076] Specifically, a preset power prediction model corresponding to the target characteristic data and the type of the target new energy source may be applied to determine the power prediction result of the target new energy source to improve the accuracy of the power prediction.

[0077] In order to improve the reliability of power prediction model training and thus improve the accuracy of applied power prediction models, such as Figure 2 As shown, in one embodiment, the new energy power prediction method based on time series decomposition further includes:

[0078] Step 001: Collect batch historical original time series and their corresponding actual powers.

[0079] Step 002: Perform time domain decomposition on each historical original time series to obtain the historical trend term, historical seasonal term and historical residual term corresponding to the historical original time series, and perform frequency domain decomposition on the historical residual term corresponding to each historical original time series to obtain the historical modal characteristics of the historical original time series.

[0080] Step 003: Perform feature screening based on the historical trend item and the historical season item corresponding to each historical original time series to obtain historical feature data of the historical original time series.

[0081] Step 004: Use the historical feature data, historical modal features and actual power corresponding to each historical original time series to train the bidirectional long short-term memory network model to obtain the preset power prediction model.

[0082] Specifically, the first historical original time series of batch historical wind farms and their corresponding actual powers can be collected; the historical feature data and actual powers corresponding to each first historical original time series are used to train the bidirectional long short-term memory network model to obtain the power prediction model corresponding to the wind farm; the second historical original time series of batch historical photovoltaic power stations and their corresponding actual powers can be collected; the historical feature data and actual powers corresponding to each second historical original time series are used to train the bidirectional long short-term memory network model to obtain the power prediction model corresponding to the photovoltaic power station.

[0083] In order to avoid errors in directly predicting residual terms and improve the accuracy of subsequent power prediction, in one embodiment, step 200 performs frequency domain decomposition on the residual terms to obtain modal features, including: applying a variational modal decomposition method to perform frequency domain decomposition on the residual terms to obtain modal features.

[0084] In order to improve the reliability of feature selection, in one embodiment, step 300 includes: applying the maximum mutual information coefficient to perform feature screening on the trend item and the season item to obtain target feature data.

[0085] In order to improve the reliability of power prediction, in one embodiment, step 400 includes:

[0086] Step 401: Apply the target characteristic data and a preset first sub-power prediction model to determine a first intermediate power prediction value.

[0087] Step 402: Apply the modal characteristics and a preset second sub-power prediction model to determine a second intermediate power prediction value.

[0088] Step 403: Determine the sum of the first intermediate power prediction value and the second intermediate power prediction value as the power prediction result of the target new energy; the preset power prediction model includes: the preset first sub-power prediction model and the preset second sub-power prediction model.

[0089] In order to improve the reliability of power prediction model training, in one embodiment, step 004 includes:

[0090] Step 041: Input the historical feature data into the first bidirectional long short-term memory network model to obtain the output result of the first bidirectional long short-term memory network model, and input the historical modal features into the second bidirectional long short-term memory network model to obtain the output result of the second bidirectional long short-term memory network model.

[0091] It can be understood that the first bidirectional long short-term memory network model can represent a bidirectional long short-term memory network model of input historical feature data, and the second bidirectional long short-term memory network model can represent a bidirectional long short-term memory network model of input historical modal features.

[0092] Step 042: Using the sum of the output results of the first bidirectional long short-term memory network model and the second bidirectional long short-term memory network model and the corresponding actual power, train to obtain the preset power prediction model; wherein the preset first sub-power prediction model is the trained first bidirectional long short-term memory network model, and the second sub-power prediction model is the trained second bidirectional long short-term memory network model.

[0093] Specifically, a loss function, such as mean square error (MSE) or mean absolute error (MAE), may be applied to measure the difference between the sum of the output results and the actual power, so as to complete the training of the preset power prediction model.

[0094] In order to further illustrate the present solution, the present application provides an application example of a new energy power prediction method based on time series decomposition, which is specifically described as follows:

[0095] Step 1: Wind-solar time domain decomposition based on STL.

[0096] First, the wind-solar power forecast is decomposed in the time domain. Time domain decomposition is a time series analysis method that can decompose complex time series into multiple relatively simple components. These components include: long-term trends, seasonal components and random noise. By decomposing the time series, we can better understand and model the complex patterns in the data and improve the accuracy of the prediction. The time domain decomposition method can not only identify and extract the main trends and cycles in the data, but also separate the random fluctuations in the data, providing a basis for subsequent frequency domain analysis and modeling.

[0097] The STL method is a powerful time series decomposition method that has been widely used in time series analysis. This method independently models the trend and seasonal components of the time series, making it more accurate in capturing long-term trends and seasonal fluctuations.

[0098] The core idea of ​​the STL method is to decompose the time series into trend term, seasonal term and residual term through local weighted regression Loess. The advantage of this method is its flexibility, which can adapt to seasonal changes of different frequencies and amplitudes. It is particularly suitable for processing wind-solar time series data with complex seasonality and nonlinear trends. The time domain decomposition steps based on STL include:

[0099] The mathematical expression of STL decomposition is:

[0100] X t = T t +S t + R t (1)

[0101] Among them, X t is the original time series of wind power and photovoltaic power, T t is the trend term, S t is the seasonal term, R t is the residual term.

[0102] Step 11: Trend item extraction:

[0103] The trend term reflects the long-term trend of wind power and photovoltaic time series. It is a relatively smooth component in the data and can usually be used to capture the basic trend of the data. The extraction of the trend term is achieved through local weighted regression (Loess). Loess is a non-parametric regression method that fits a low-order polynomial near each data point to obtain a smooth trend curve. Its mathematical expression is:

[0104] T t =Loess(X t ) (2)

[0105] Step 12: Seasonal term extraction:

[0106] The seasonal term captures the periodic fluctuations in the time series. It is a highly repetitive part that can reflect seasonal or periodic characteristics. The specific method is to perform periodic fitting on the detrended data to obtain the seasonal term S t The extraction process of seasonal terms can be expressed as:

[0107]

[0108] Where n is the cycle length, X t -T t Represents the data after trending. Extraction of seasonal components usually involves segmenting the data, with each segment corresponding to a complete cycle. Seasonal fluctuations can be effectively extracted by calculating the average value of each segment or fitting it with a periodic model.

[0109] Step 13: Residual extraction:

[0110] The residual term contains random noise and short-term fluctuations in the time series, which is the part left after decomposition that cannot be explained by trend terms and seasonal terms. In practical applications, the analysis and processing of residual terms are of great significance to improving the accuracy of the prediction model. The extraction process of residual terms can be expressed as:

[0111] R t =X t -T t -S t (4)

[0112] Step 2: Wind-solar frequency domain decomposition based on variational mode decomposition.

[0113] After time domain decomposition, although the trend and seasonal components in the time series are separated, the residual term still contains a large number of high-frequency components and random noise, which are strong non-stationary components. Direct prediction of this component often has large errors. At the same time, these high-frequency components may contain useful information and need to be further extracted and analyzed. Frequency domain decomposition is an effective method that can better capture the different frequency components in the time series by converting the signal to the frequency domain.

[0114] VMD is an adaptive, non-recursive modal decomposition method that can decompose complex signals into multiple limited-bandwidth modes with different center frequencies and has strong anti-noise capabilities. Compared with traditional empirical mode decomposition (EMD), VMD has better stability and adaptability and can more accurately capture the frequency information in the signal.

[0115] The core idea of ​​the VMD method is to decompose the signal into several modes by solving a variational problem. Each mode corresponds to a center frequency and has a minimum bandwidth. VMD simultaneously updates the modal function and its corresponding center frequency through an iterative optimization method until convergence.

[0116] The mathematical model of VMD is:

[0117]

[0118] in, u k is the kth mode function, ω k is the corresponding center frequency, and δ(t) is the Dirac function.

[0119] The VMD decomposition steps include:

[0120] Through variational mode decomposition, the signal R t is decomposed into several modes uk (t), each mode has a narrowband characteristic. The goal of VMD is to find the optimal mode function u k and the corresponding center frequency ω k , so that the bandwidth of each mode is minimized.

[0121] During the iterative optimization process, the modal function and center frequency are first initialized, and then the modal function and center frequency are updated alternately until convergence. The specific optimization process includes:

[0122] Step 21: Initialization: Set the number of modes K and the initial mode function u k and the center frequency ω k .

[0123] Step 22: Modal function update: At a fixed center frequency ω k In the case of u k :

[0124]

[0125] Step 23: Center Frequency Update: In the Fixed Mode Function u k In the case of k :

[0126]

[0127] Step 24: Initialization: Determine whether the modal function and center frequency have converged. If not, return to step 22 to continue iterating.

[0128] By performing frequency domain decomposition on the residual terms of historical wind power and photovoltaic data, useful information can be further extracted and the accuracy of power prediction can be improved.

[0129] Step 3: Feature screening based on the maximum mutual information coefficient.

[0130] When making short-term power forecasts for wind power and photovoltaic power generation, in addition to considering the data after time domain and frequency domain decomposition, it is also necessary to select appropriate features to input into the forecasting model. Feature selection is a key step in machine learning, which directly affects the performance and forecasting accuracy of the model. For wind power and photovoltaic power generation data, common features include: historical power data, meteorological data (such as temperature, wind speed, light intensity, etc.) and other related factors (such as radiation, sun angle, weather conditions, etc.).

[0131] Maximum Information Coefficient (MIC) is a non-parametric statistical method used to measure the mutual information between two variables. Unlike the traditional correlation coefficient, MIC can capture linear and nonlinear relationships and is a more general correlation analysis tool. The core idea of ​​the MIC method is to find the maximum amount of information between variables through grid search to measure their correlation. Specifically, the MIC method can capture various relationships between variables, including linear, nonlinear, periodic, etc., and is applicable to various complex data structures. The MIC feature screening steps are as follows:

[0132] Step 31: Data collection: For wind power forecasting, the collected features include: wind speed, temperature, historical wind power and weather data. For photovoltaic forecasting, the collected features include: radiation, sun angle, historical photovoltaic power and weather conditions.

[0133] Step 32: Mutual Information Calculation: The mutual information between each pair of features and the target variable is calculated using the mutual information formula.

[0134]

[0135] Where I(X;Y) is the mutual information of X and Y, X is the feature to be screened, which may include trend terms and seasonal terms, Y is the power generation of wind power or photovoltaic power, x and y are the values ​​of feature X and power generation Y respectively. |X| and |Y| are the value ranges of X and Y respectively. p(x,y) is the joint probability distribution, and p(x) and p(y) are the marginal probability distributions.

[0136] Step 33: MIC calculation: Use the MIC formula to calculate the maximum mutual information coefficient between each pair of features and the target variable.

[0137]

[0138] Among them, X represents the feature to be screened, such as wind speed, temperature, radiation, sun angle, etc. Y represents the target variable, that is, the power generation of wind power or photovoltaic power.

[0139] Step 34: Feature selection: Select the features with the highest correlation with the target variable as the input of the prediction model. For example, for wind power prediction, you can select wind speed, temperature and other highly correlated features. For photovoltaic prediction, you can select radiation, sun angle and other highly correlated features.

[0140] In summary, feature screening plays a key role in short-term power forecasting of wind power and photovoltaic power generation. By using the MIC method, the features with the highest correlation with the target variable can be effectively screened out, significantly improving the performance and accuracy of the prediction model. Steps 1 to 3 can realize feature data processing, decouple wind-photovoltaic power data, and facilitate the mining of wind-photovoltaic features.

[0141] Step 4: Prediction model and framework based on BiLSTM.

[0142] After completing the time domain decomposition and frequency domain decomposition, as well as feature screening, it is necessary to build an effective prediction model to predict the short-term power of wind power and photovoltaic power generation. Currently, deep learning methods perform well in processing time series prediction tasks. In particular, the BiLSTM model combines LSTM and the bidirectional structural information of data. It can mine the intrinsic laws of time series data in the past and future moments, and performs well in time series prediction tasks.

[0143] The BiLSTM model introduces a bidirectional time series information transmission structure. While analyzing forward information, it also mines future backward information, which increases the model's perception of the features before and after the time series, thereby better establishing long-series time dependencies. Wind power and photovoltaic power are greatly affected by meteorology. When meteorological data changes, BiLSTM combines the related information before and after the time series, showing a higher sensitivity to the changing data of the time series, thus further improving the time series prediction effect of photovoltaic and wind power.

[0144] The basic architecture of the BiLSTM model includes forward LSTM and backward LSTM, which can simultaneously mine the temporal characteristics of both forward and reverse data in the sequence. The forward calculation process of the BiLSTM neural network is:

[0145] h ft =H(x t ,h ft-1 ) (10)

[0146] Among them, h ft is the forward hidden layer state, and H is the LSTM calculation process.

[0147] The backward calculation process of the BiLSTM neural network is:

[0148] h bt =H(x t ,h bt-1 ) (11)

[0149] Among them, h bt is the backward hidden layer state, and H is the LSTM calculation process.

[0150] The final output of the BiLSTM model is:

[0151] h t =w ft h ft +w bt h bt (12)

[0152] Among them, w ft is the forward hidden layer output weight, w bt Output weights for the backward hidden layer.

[0153] The LSTM network contains three logic units: input gate, forget gate, and output gate, which control the output of the memory unit. The specific calculation process is:

[0154] i t =σ(W i (x t ,h t-1 )+b i ) (13)

[0155] f t =σ(W f (x t ,h t-1 )+b f ) (14)

[0156] o t =σ(W o (x t ,h t-1 )+b o ) (15)

[0157] C t =tanh(W c (x t ,h t-1 )+b c ) (16)

[0158] c t =i t ·C t +f t ·c t-1 (17)

[0159] h ft / bt =o t ·tanh(c t ) (18)

[0160] Among them, x t is the input, and tanh is the activation function, W and b are the corresponding weights and biases, c tIt represents the state of the memory storage unit and is the core of LSTM.

[0161] 1) Overall framework for wind and solar forecasting: In order to achieve short-term power forecasting for wind and photovoltaic power generation, this application integrates time domain decomposition, frequency domain decomposition, maximum mutual information coefficient (MIC) and BiLSTM model into a unified forecasting framework. First, the original wind power and photovoltaic data are decomposed in the time domain to extract trend terms, seasonal terms and residual terms; then the residual terms are subjected to variational mode decomposition (VMD) to extract modal functions of different frequencies; then, the maximum mutual information coefficient (MIC) method is used to screen features that are highly correlated with power generation; finally, the screened features are input into the BiLSTM model for training and forecasting.

[0162] 2) Prediction error evaluation index: In order to evaluate the performance of the prediction model, this application uses a variety of typical prediction error evaluation indexes, including root mean square error (RMSE) and mean absolute error (MAE). These indicators are defined as follows:

[0163] Among them, y i is the actual value, is the predicted value, and n is the number of samples.

[0164] Root Mean Square Error (RMSE):

[0165]

[0166] Mean Absolute Error (MAE):

[0167]

[0168] In order to verify the effectiveness of the new energy power prediction method based on time series decomposition provided in the embodiment of the present application, the following three groups of prediction cases are set for experiment:

[0169] Case 1: The BiLSTM model is used to input the relevant features selected by MIC and directly predict the undecomposed original wind and solar power data.

[0170] Case 2: The BiLSTM model is used to input the relevant features selected by MIC to predict the seasonal term, trend term, and residual term of the wind-solar time domain decomposition.

[0171] Case 3: BiLSTM model is used to input the relevant features selected by MIC to predict the seasonal term and trend term of wind and solar time domain decomposition. VMD is used to decompose the residual term in the frequency domain, and the obtained modal features are used as the prediction input of the residual term.

[0172] Table 1

[0173] experiment RMSE / kW MAE / kW Case 1 Photovoltaic Forecast 3.28 2.05 Case 2 Photovoltaic Forecast 2.56 1.63 Case 3 Photovoltaic Forecast 0.92 0.58 Case 1 Wind Power Forecast 124.27 98.37 Case 2 Wind Power Forecast 89.79 67.58 Case 3 Wind Power Forecast 67.50 48.16

[0174] Figure 3 The following is a schematic diagram comparing the photovoltaic power prediction results of the above different experimental cases. Figure 4 This is a schematic diagram comparing the wind power prediction results of the above different experimental cases. Figure 3 and Figure 4 The horizontal axis represents time, and the vertical axis represents power. Figure 3 and Figure 4 Curve 0 in the figure represents the real power curve, and curves 1 to 3 represent the power prediction curves of case 1 to case 3 respectively. Table 1 shows the root mean square error RMSE and mean absolute error MAE of the wind and solar prediction results of the above different experimental cases. Figure 3 and Figure 4 It can be seen that the photovoltaic and wind power of Case 3 best fit the original wind-solar power curve, and the average RMSE of the photovoltaic and wind power prediction errors of Case 1 is 3.28 and 124.27, and the average MAE is 2.05 and 98.37. Since the power base of photovoltaic power is small, the prediction error is smaller than that of wind power. In addition, it can be seen from Table 1 that the wind-solar prediction error of Case 2 is lower than that of Case 1, indicating that decomposing the wind-solar power data into three time domain terms can enable BiLSTM to better learn the change law of load on the time scale. Finally, it can be obtained that the RMSE of photovoltaic and wind power prediction errors in Case 3 reaches the minimum values ​​of 0.92 and 67.50, and the MAE is 0.58 and 48.16. Compared with Case 2, it shows that only using the historical values ​​of the residual components as input features, it is difficult for the prediction model to directly learn the variation law of the residual terms with strong randomness. Case 3 performs VMD decomposition on the residual terms, and extracts the modal features of different frequencies as the input of the residual term prediction model, which can simplify the complex residual fluctuation situation and improve the residual terms and the overall wind and solar power prediction accuracy. It has been verified that the time-frequency decomposition method proposed in the embodiment of the present application can effectively learn the fluctuation law in the wind and solar time series, thereby improving the prediction accuracy of wind and solar power.

[0175] From the software level, in order to improve the reliability of wind and solar power prediction on the basis of ensuring the efficiency of wind and solar power prediction, the present application provides an embodiment of a new energy power prediction device based on time series decomposition for realizing all or part of the contents of the new energy power prediction method based on time series decomposition, see Figure 5 The new energy power prediction device based on time series decomposition specifically includes the following contents:

[0176] Acquisition module 01, used to acquire the original time series of the target new energy;

[0177] The first decomposition module 02 is used to perform time domain decomposition on the original time series to obtain trend terms, seasonal terms and residual terms, and perform frequency domain decomposition on the residual terms to obtain modal features;

[0178] The first screening module 03 is used to perform feature screening based on the trend item and the season item to obtain target feature data;

[0179] Prediction module 04 is used to apply the target feature data, modal features and a preset power prediction model to determine the power prediction result of the target new energy. The preset power prediction model is pre-trained based on a bidirectional long short-term memory network model of batch historical original time series and their respective corresponding actual powers.

[0180] like Figure 6 As shown, in one embodiment, the new energy power prediction device based on time series decomposition further includes:

[0181] Collection module 05, used to collect batch historical original time series and their corresponding actual power;

[0182] The second decomposition module 06 is used to perform time domain decomposition on each historical original time series to obtain the historical trend term, historical season term and historical residual term corresponding to the historical original time series, and perform frequency domain decomposition on the historical residual term corresponding to each historical original time series to obtain the historical modal features of the historical original time series;

[0183] The second screening module 07 is used to perform feature screening based on the historical trend item and the historical season item corresponding to each historical original time series to obtain historical feature data of the historical original time series;

[0184] The training module 08 is used to train the bidirectional long short-term memory network model using the historical feature data, historical modal features and actual power corresponding to each historical original time series to obtain the preset power prediction model.

[0185] In one embodiment, the first decomposition module comprises:

[0186] The frequency domain decomposition unit is used to apply the variational mode decomposition method to perform frequency domain decomposition on the residual term to obtain the mode feature.

[0187] In one embodiment, the first screening module includes:

[0188] The screening unit is used to apply the maximum mutual information coefficient to perform feature screening on the trend item and the seasonal item to obtain target feature data.

[0189] In one embodiment, the prediction module includes:

[0190] A first determining unit, configured to determine a first intermediate power prediction value by applying the target characteristic data and a preset first sub-power prediction model;

[0191] A second determining unit, configured to determine a second intermediate power prediction value by applying the modal feature and a preset second sub-power prediction model;

[0192] Determine the sum of the first intermediate power prediction value and the second intermediate power prediction value as the power prediction result of the target new energy;

[0193] The prediction unit is used for the preset power prediction model including: the preset first sub-power prediction model and the preset second sub-power prediction model.

[0194] In one embodiment, the training module includes:

[0195] An input unit, used to input the historical feature data into a first bidirectional long short-term memory network model to obtain an output result of the first bidirectional long short-term memory network model, and input the historical modal features into a second bidirectional long short-term memory network model to obtain an output result of the second bidirectional long short-term memory network model;

[0196] A training unit, configured to train the preset power prediction model using the sum of the output results of the first bidirectional long short-term memory network model and the second bidirectional long short-term memory network model and the corresponding actual power;

[0197] Among them, the preset first sub-power prediction model is a trained first bidirectional long short-term memory network model, and the second sub-power prediction model is a trained second bidirectional long short-term memory network model.

[0198] The embodiment of the new energy power prediction device based on time series decomposition provided in this specification can be specifically used to execute the processing flow of the embodiment of the new energy power prediction method based on time series decomposition mentioned above. Its functions will not be repeated here, and reference can be made to the detailed description of the embodiment of the new energy power prediction method based on time series decomposition mentioned above.

[0199] From the hardware level, in order to improve the reliability of wind and solar power prediction on the basis of ensuring the efficiency of wind and solar power prediction, the present application provides an embodiment of an electronic device for implementing all or part of the contents of the new energy power prediction method based on time series decomposition. The electronic device specifically includes the following contents:

[0200] Processor, memory, communication interface and bus; wherein the processor, memory and communication interface communicate with each other through the bus; the communication interface is used to realize information transmission between the new energy power prediction device based on time series decomposition and related devices such as user terminals; the electronic device can be a desktop computer, a tablet computer and a mobile terminal, etc., but the present embodiment is not limited thereto. In the present embodiment, the electronic device can be implemented with reference to the embodiment for realizing the new energy power prediction method based on time series decomposition and the embodiment for realizing the new energy power prediction device based on time series decomposition, and the contents thereof are incorporated herein, and the repeated parts are not repeated.

[0201] Figure 7 FIG. 9 is a schematic block diagram of the system structure of the electronic device 9600 according to an embodiment of the present application. Figure 7 As shown, the electronic device 9600 may include a central processor 9100 and a memory 9140; the memory 9140 is coupled to the central processor 9100. It is worth noting that Figure 7 is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.

[0202] In one or more embodiments of the present application, the power prediction function of wind and solar power can be integrated into the central processor 9100. The central processor 9100 can be configured to perform the following control:

[0203] Step 100: Obtain the original time series of the target new energy.

[0204] Step 200: Decompose the original time series in the time domain to obtain trend terms, seasonal terms and residual terms, and decompose the residual terms in the frequency domain to obtain modal features.

[0205] Step 300: Perform feature screening based on the trend item and the season item to obtain target feature data.

[0206] Step 400: Apply the target feature data, modal features and a preset power prediction model to determine the power prediction result of the target new energy. The preset power prediction model is pre-trained based on a bidirectional long short-term memory network model based on batch historical original time series and their respective corresponding actual powers.

[0207] It can be seen from the above description that the electronic device provided by the embodiment of the present application can improve the reliability of wind and solar power prediction while ensuring the efficiency of wind and solar power prediction.

[0208] In another embodiment, the new energy power prediction device based on time series decomposition can be configured separately from the central processing unit 9100. For example, the new energy power prediction device based on time series decomposition can be configured as a chip connected to the central processing unit 9100, and the power prediction function of wind and solar power can be realized through the control of the central processing unit.

[0209] like Figure 7 As shown, the electronic device 9600 may also include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Figure 7 In addition, the electronic device 9600 may also include Figure 7 For components not shown, reference may be made to the prior art.

[0210] like Figure 7 As shown, the central processing unit 9100 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0211] The memory 9140 may be, for example, one or more of a cache, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory or other suitable devices. The above-mentioned information related to the failure may be stored, and a program for executing the relevant information may also be stored. The CPU 9100 may execute the program stored in the memory 9140 to implement information storage or processing, etc.

[0212] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 is, for example, a key or a touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display display objects such as images and texts. The display may be, for example, an LCD display, but is not limited thereto.

[0213] The memory 9140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It may also be a memory that saves information even when the power is off, can be selectively erased, and is provided with more data, examples of which are sometimes referred to as EPROMs, etc. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142, which is used to store application programs and function programs or processes for executing the operation of the electronic device 9600 through the central processor 9100.

[0214] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0215] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module (transmitter / receiver) is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.

[0216] Based on different communication technologies, multiple communication modules 9110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module and / or a wireless LAN module, etc. The communication module (transmitter / receiver) 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide an audio output via the speaker 9131 and receive an audio input from the microphone 9132, thereby realizing a common telecommunication function. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 9130 is also coupled to the central processor 9100, so that recording can be performed on the local machine through the microphone 9132, and the sound stored on the local machine can be played through the speaker 9131.

[0217] It can be seen from the above description that the electronic device provided by the embodiment of the present application can improve the reliability of wind and solar power prediction while ensuring the efficiency of wind and solar power prediction.

[0218] The embodiment of the present application also provides a computer-readable storage medium capable of implementing all the steps in the new energy power prediction method based on time series decomposition in the above embodiment. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, all the steps of the new energy power prediction method based on time series decomposition in the above embodiment are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0219] Step 100: Obtain the original time series of the target new energy.

[0220] Step 200: Decompose the original time series in the time domain to obtain trend terms, seasonal terms and residual terms, and decompose the residual terms in the frequency domain to obtain modal features.

[0221] Step 300: Perform feature screening based on the trend item and the season item to obtain target feature data.

[0222] Step 400: Apply the target feature data, modal features and a preset power prediction model to determine the power prediction result of the target new energy. The preset power prediction model is pre-trained based on a bidirectional long short-term memory network model based on batch historical original time series and their respective corresponding actual powers.

[0223] It can be seen from the above description that the computer-readable storage medium provided in the embodiment of the present application can improve the reliability of wind and solar power prediction while ensuring the efficiency of wind and solar power prediction.

[0224] In this application, each embodiment of the above method is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. For relevant parts, refer to the partial description of the method embodiment.

[0225] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0226] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0227] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0228] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0229] Specific embodiments are used in this application to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core idea of ​​this application. At the same time, for those skilled in the art, according to the idea of ​​this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on this application.

Claims

1. A new energy power prediction method based on time series decomposition, characterized in that: include: Obtain the original time series of the target new energy; Decomposing the original time series in the time domain to obtain trend terms, seasonal terms and residual terms, and decomposing the residual terms in the frequency domain to obtain modal features; Perform feature screening based on the trend item and the season item to obtain target feature data; The target feature data, modal features and a preset power prediction model are applied to determine the power prediction result of the target new energy. The preset power prediction model is obtained by pre-training a bidirectional long short-term memory network model based on batch historical original time series and their respective corresponding actual powers.

2. The new energy power prediction method based on time series decomposition according to claim 1 is characterized in that: Also includes: Collect batch historical raw time series and their corresponding actual power; Perform time domain decomposition on each historical original time series to obtain the historical trend term, historical seasonal term and historical residual term corresponding to the historical original time series, and perform frequency domain decomposition on the historical residual term corresponding to each historical original time series to obtain the historical modal characteristics of the historical original time series; Perform feature screening based on the historical trend item and the historical season item corresponding to each historical original time series to obtain historical feature data of the historical original time series; The bidirectional long short-term memory network model is trained using the historical feature data, historical modal features and actual power corresponding to each historical original time series to obtain the preset power prediction model.

3. The new energy power prediction method based on time series decomposition according to claim 1 is characterized in that: The performing frequency domain decomposition on the residual term to obtain modal features includes: The variational mode decomposition method is applied to decompose the residual term in the frequency domain to obtain the mode characteristics.

4. The new energy power prediction method based on time series decomposition according to claim 1 is characterized in that: The feature screening based on the trend item and the season item to obtain target feature data includes: The maximum mutual information coefficient is applied to perform feature screening on the trend item and the seasonal item to obtain target feature data.

5. The new energy power prediction method based on time series decomposition according to claim 1 is characterized in that: The applying the target characteristic data, the modal characteristics and the preset power prediction model to determine the power prediction result of the target new energy source includes: Applying the target characteristic data and a preset first sub-power prediction model to determine a first intermediate power prediction value; Applying the modal characteristics and a preset second sub-power prediction model to determine a second intermediate power prediction value; Determine the sum of the first intermediate power prediction value and the second intermediate power prediction value as the power prediction result of the target new energy; The preset power prediction model includes: the preset first sub-power prediction model and the preset second sub-power prediction model.

6. The new energy power prediction method based on time series decomposition according to claim 2 is characterized in that: The application of the historical feature data, historical modal features and actual power corresponding to each historical original time series to train the bidirectional long short-term memory network model to obtain the preset power prediction model includes: Input the historical feature data into a first bidirectional long short-term memory network model to obtain an output result of the first bidirectional long short-term memory network model, and input the historical modal features into a second bidirectional long short-term memory network model to obtain an output result of the second bidirectional long short-term memory network model; The preset power prediction model is obtained by training using the sum of the output results of the first bidirectional long short-term memory network model and the second bidirectional long short-term memory network model and the corresponding actual power; Among them, the preset first sub-power prediction model is a trained first bidirectional long short-term memory network model, and the second sub-power prediction model is a trained second bidirectional long short-term memory network model.

7. A new energy power prediction device based on time series decomposition, characterized in that: include: An acquisition module is used to obtain the original time series of the target new energy; A first decomposition module is used to perform time domain decomposition on the original time series to obtain trend terms, seasonal terms and residual terms, and perform frequency domain decomposition on the residual terms to obtain modal features; A first screening module, used for performing feature screening based on the trend item and the season item to obtain target feature data; A prediction module is used to apply the target feature data, modal features and a preset power prediction model to determine the power prediction result of the target new energy. The preset power prediction model is obtained by pre-training a bidirectional long short-term memory network model based on batch historical original time series and their respective corresponding actual powers.

8. The new energy power prediction device based on time series decomposition according to claim 7 is characterized in that: Also includes: The acquisition module is used to collect batch historical original time series and their corresponding actual power; The second decomposition module is used to perform time domain decomposition on each historical original time series to obtain the historical trend term, historical seasonal term and historical residual term corresponding to the historical original time series, and perform frequency domain decomposition on the historical residual term corresponding to each historical original time series to obtain the historical modal features of the historical original time series; A second screening module is used to perform feature screening based on the historical trend item and the historical season item corresponding to each historical original time series to obtain historical feature data of the historical original time series; The training module is used to train the bidirectional long short-term memory network model by applying the historical feature data, historical modal features and actual power corresponding to each historical original time series to obtain the preset power prediction model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the new energy power prediction method based on time series decomposition according to any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instructions are executed by the processor, the new energy power prediction method based on time series decomposition according to any one of claims 1 to 6 is implemented.

Images