Frequency Modulation Instruction Prediction Method and System for Supercapacitor-Coupled Lithium Battery Energy Storage

Abstract

The present application relates to the technical field of instruction prediction, and discloses a frequency modulation instruction prediction method and system for supercapacitor-coupled lithium battery energy storage. The method includes: decomposing a standardized frequency modulation sequence through a smoothing function to obtain an approximately integrable quantity and a non-integrable quantity; performing trigonometric function periodic conversion processing on the approximately integrable quantity to obtain a periodic feature sequence, which is composed of sine and cosine functions with multiple different frequencies; performing linear combination transformation processing on the periodic feature sequence and the non-integrable quantity to obtain a combined feature sequence; training the combined feature sequence through a deep neural network to obtain a frequency modulation instruction prediction model; predicting real-time frequency modulation instruction data through the frequency modulation instruction prediction model to obtain a prediction result, and performing anti-normalization processing on the prediction result to obtain a frequency modulation instruction prediction value. The present application improves the efficiency and accuracy of frequency modulation instruction prediction for supercapacitor-coupled lithium battery energy storage.

Description

Technical Field

[0001] This application relates to the field of instruction prediction, and particularly to a frequency modulation instruction prediction method and system for supercapacitor-coupled lithium battery energy storage. Background Art

[0002] In the power system, thermal power units are the main frequency modulation power sources and undertake the important task of grid frequency regulation. With the large-scale integration of new energy into the grid and the increasing environmental protection requirements, the demand for frequency modulation services in the grid is continuously increasing. When traditional thermal power units perform frequency regulation, they need to frequently change the output of the units. This operation mode will lead to problems such as increased coal consumption of the units, aggravated equipment wear, and shortened operation life. To solve this problem, introducing an energy storage system as an auxiliary frequency modulation device into thermal power plants has become an important technical solution. Among them, a hybrid energy storage system composed of supercapacitors and lithium batteries is widely used in the field of power frequency modulation due to its advantages such as high power density, large energy density, and fast response speed.

[0003] However, the existing energy storage system auxiliary frequency modulation technology has the problem of insufficient prediction accuracy. Due to the characteristics of strong nonlinearity and non-regularity of frequency modulation instructions, traditional prediction methods often directly input the original frequency modulation instructions into the prediction model without fully considering the mathematical characteristics of the signals, resulting in large errors in the prediction results. This lack of prediction accuracy will affect the power distribution decision of the energy storage system, reduce the frequency modulation performance of the system, and cannot give full play to the advantages of the hybrid energy storage system. Summary of the Invention

[0004] This application provides a frequency modulation instruction prediction method and system for supercapacitor-coupled lithium battery energy storage, which is used to improve the efficiency and accuracy of frequency modulation instruction prediction for supercapacitor-coupled lithium battery energy storage.

[0005] In a first aspect, the present application provides a frequency modulation command prediction method for a supercapacitor-coupled lithium battery energy storage system. The frequency modulation command prediction method for a supercapacitor-coupled lithium battery energy storage system includes: performing data cleaning and normalization processing on the original frequency modulation command sequence to obtain a standardized frequency modulation sequence; decomposing the standardized frequency modulation sequence through a smoothing function to obtain an approximately integrable quantity and a non-integrable quantity; performing trigonometric function periodic conversion processing on the approximately integrable quantity to obtain a periodic feature sequence, where the periodic feature sequence is composed of sine and cosine functions with multiple different frequencies; performing linear combination transformation processing on the periodic feature sequence and the non-integrable quantity through weight coefficients to obtain a combined feature sequence; training the combined feature sequence through a deep neural network to obtain a frequency modulation command prediction model, where the frequency modulation command prediction model includes a three-layer convolutional neural network, a bidirectional long short-term memory network, and a fully connected layer; predicting real-time frequency modulation command data through the frequency modulation command prediction model to obtain a prediction result, and performing anti-normalization processing on the prediction result to obtain a frequency modulation command prediction value.

[0006] In a second aspect, the present application provides a frequency modulation command prediction system for a supercapacitor-coupled lithium battery energy storage system. The frequency modulation command prediction system for a supercapacitor-coupled lithium battery energy storage system includes:

[0007] A processing module, configured to perform data cleaning and normalization processing on the original frequency modulation command sequence to obtain a standardized frequency modulation sequence;

[0008] A decomposition module, configured to decompose the standardized frequency modulation sequence through a smoothing function to obtain an approximately integrable quantity and a non-integrable quantity;

[0009] A conversion module, configured to perform trigonometric function periodic conversion processing on the approximately integrable quantity to obtain a periodic feature sequence, where the periodic feature sequence is composed of sine and cosine functions with multiple different frequencies;

[0010] A transformation module, configured to perform linear combination transformation processing on the periodic feature sequence and the non-integrable quantity through weight coefficients to obtain a combined feature sequence;

[0011] A training module, configured to train the combined feature sequence through a deep neural network to obtain a frequency modulation command prediction model, where the frequency modulation command prediction model includes a three-layer convolutional neural network, a bidirectional long short-term memory network, and a fully connected layer;

[0012] A prediction module, configured to predict real-time frequency modulation command data through the frequency modulation command prediction model to obtain a prediction result, and perform anti-normalization processing on the prediction result to obtain a frequency modulation command prediction value.

[0013] In the technical solution provided by this application, through data cleaning and normalization processing of the original frequency modulation instruction sequence, missing values, outliers, and noise interference in the data are effectively removed, ensuring the reliability of data quality. By decomposing the standardized frequency modulation sequence using a smoothing function, the complex non-linear sequence is decomposed into two parts: an approximately integrable quantity and a non-integrable quantity, which is convenient for separate processing and improves the prediction accuracy. The approximately integrable quantity is subjected to trigonometric function periodic conversion processing to convert the irregular time series into a combination of periodic functions with clear mathematical characteristics, enhancing the predictability of the signal. Through linear combination transformation processing of the periodic feature sequence and the non-integrable quantity using weight coefficients, the organic combination of periodic changes and random fluctuations is achieved, enabling the prediction model to capture both the regular and random characteristics of the signal simultaneously. A deep neural network is used to train the combined feature sequence, making full use of the advantages of the convolutional neural network to extract local features and the long short-term memory network to capture long-term dependence relationships, and a model with powerful prediction ability is constructed. In the real-time prediction stage, through preprocessing and anti-normalization processing, the accuracy and practicality of the prediction results are ensured. The overall solution significantly improves the prediction accuracy of the frequency modulation instruction through the system design of data preprocessing, feature extraction, model training, and prediction output. Experimental verification shows that the prediction RMSE of this method for frequency modulation instruction 1 and frequency modulation instruction 2 are 1.2 and 3.3 respectively, showing a significant improvement compared to 9.9 and 8.7 of the traditional model, providing a reliable decision-making basis for the efficient operation of the supercapacitor and lithium battery hybrid energy storage system. This improvement in prediction accuracy helps to optimize the power distribution strategy of the energy storage system, improve the frequency modulation performance of the system, extend the service life of the equipment, and reduce the operating cost. Description of the Drawings

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0015] Figure 1 Schematic diagram of an embodiment of the frequency modulation instruction prediction method for supercapacitor-coupled lithium battery energy storage in an embodiment of this application;

[0016] Figure 2 Schematic diagram of an embodiment of the frequency modulation instruction prediction system for supercapacitor-coupled lithium battery energy storage in an embodiment of this application. Detailed Embodiments

[0017] The embodiments of the present application provide a method and system for predicting frequency modulation commands for supercapacitor-coupled lithium battery energy storage. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned drawings of the present application are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figure 1 One embodiment of the method for predicting frequency modulation commands for supercapacitor-coupled lithium battery energy storage in the embodiments of the present application includes:

[0019] Step S101: Perform data cleaning and normalization processing on the original frequency modulation command sequence to obtain a standardized frequency modulation sequence;

[0020] Step S102: Decompose the standardized frequency modulation sequence through a smoothing function to obtain an approximately integrable quantity and a non-integrable quantity;

[0021] Step S103: Perform trigonometric periodic conversion processing on the approximately integrable quantity to obtain a periodic feature sequence, where the periodic feature sequence consists of sine and cosine functions of multiple different frequencies;

[0022] Step S104: Perform linear combination transformation processing on the periodic feature sequence and the non-integrable quantity through weight coefficients to obtain a combined feature sequence;

[0023] Step S105: Perform training processing on the combined feature sequence through a deep neural network to obtain a frequency modulation command prediction model. The frequency modulation command prediction model includes a three-layer convolutional neural network, a bidirectional long short-term memory network, and a fully connected layer;

[0024] Step S106: Perform prediction processing on the real-time frequency modulation command data through the frequency modulation command prediction model to obtain a prediction result, and perform anti-normalization processing on the prediction result to obtain a frequency modulation command prediction value.

[0025] It can be understood that the execution entity of the present application can be a frequency modulation command prediction system for supercapacitor-coupled lithium battery energy storage, or a terminal or a server. Specifically, it is not limited here. The embodiments of the present application will be described by taking the server as the execution entity as an example.

[0026] Specifically, data cleaning and normalization are performed on the original frequency modulation instruction sequence. This process includes key steps such as missing value detection, outlier detection, time window partitioning, amplitude normalization, and frequency domain filtering. In the missing value detection step, by scanning the data points in the original frequency modulation instruction sequence, the positions with empty or unreasonable values are marked to obtain the missing value position information. Linear interpolation is used to repair the data at these positions to achieve the integrity of the sequence. Subsequently, outlier detection is carried out. Statistical methods are used to identify the data points that deviate significantly from the overall distribution, which are marked as outliers and replaced with the median to ensure the rationality of the data. Then, the complete sequence is partitioned into time windows. A fixed time length, such as 5 minutes per window, is set, and a 75% overlap rate is used for sliding to increase the data correlation. For the partitioned sequence samples, the amplitude range is calculated, and min-max normalization is performed to map the values to the 0-1 interval. Finally, the frequency characteristics of the sequence are analyzed through Fourier transform, and a band-pass filter is designed to remove high-frequency noise to obtain the standardized frequency modulation sequence. After obtaining the standardized frequency modulation sequence, a smoothing function is used to decompose it. The purpose is to divide the sequence into an integrable part and a non-integrable part. Specifically, a Gaussian kernel function is used as the smoothing tool, and by adjusting the window length parameter, the smoothing degree of the sequence is controlled. The parameters of the Gaussian kernel function are adaptively determined by the fluctuation characteristics of the sequence to ensure the rationality of the smoothing effect. The smoothed sequence is used as the approximate integrable quantity, and the difference between the original sequence and the approximate integrable quantity forms the non-integrable quantity. This decomposition method effectively distinguishes the main trend and random fluctuation components of the sequence.

[0027] For the approximate integrable quantity obtained by decomposition, trigonometric periodic conversion processing is required. First, the basic period of the sequence is estimated using the autocorrelation analysis method to find the main period characteristics of the signal. Then, the sequence is segmented according to the period parameters for subsequent harmonic analysis. The segmented sequence is subjected to harmonic decomposition to extract the fundamental wave and each harmonic component. Sine and cosine functions are used to fit these components to obtain the corresponding amplitude and phase coefficients. Finally, the individual frequency components are superimposed according to the coefficient magnitudes to reconstruct the periodic characteristic sequence. To comprehensively utilize the information of the periodic characteristic sequence and the non-integrable quantity, linear combination transformation processing is required. First, the importance of the periodic characteristic sequence is evaluated, and its contribution degree in the original sequence is calculated to obtain the corresponding weight coefficient. At the same time, the fluctuation characteristics of the non-integrable quantity are analyzed, and the weight coefficient is determined according to its statistical characteristics. The two parts of the sequence are multiplied by the corresponding weight coefficients and then linearly combined to obtain a combined sequence that retains both periodicity and random fluctuation characteristics.

[0028] The processing of the combined feature sequence is trained using a deep neural network, and the network structure includes a convolutional layer, a recurrent layer, and a fully connected layer. First, the sequence is divided into a training set and a validation set, and an appropriate batch size is set for data loading. Local features are extracted through a three-layer convolutional neural network, and the convolutional kernel size and number of channels in each layer are set according to the data features. Then, a bidirectional long short-term memory network is used to process the temporal dependence relationship and capture the long-term and short-term change rules. Finally, the features are mapped to the prediction space through the fully connected layer to obtain a prediction model.

[0029] Preprocess the real-time collected frequency modulation command data to make its format and features consistent with the training data. Input the preprocessed data into the prediction model, and obtain preliminary prediction values through forward calculation. Conduct a validity constraint check on the prediction values to ensure that they meet the physical constraint conditions. Finally, through inverse normalization transformation, restore the prediction results to the original numerical range to obtain the final frequency modulation command prediction value.

[0030] Taking an actual frequency modulation scenario of a thermal power plant as an example, in the original frequency modulation command sequence collected by a certain thermal power unit within one hour, one data point is recorded every 5 seconds, for a total of 720 data points. Through data cleaning, it is found that there are 8 missing values and 12 outliers, which are corrected by linear interpolation and median replacement respectively. Using a 5-minute window length and a sliding step of 75 seconds for data segmentation, 48 sample sequences are obtained. Normalize these sequences, and the original amplitude range from -50MW to +50MW is mapped to the 0-1 interval. The approximate integrable quantity obtained through smooth decomposition shows obvious 15-minute periodic characteristics, and the standard deviation of the non-integrable quantity is 0.15. After periodic transformation, 3 main frequency components are obtained, with amplitudes of 0.8, 0.5, and 0.3 respectively. After weight optimization, the combined weight ratio of the periodic feature sequence and the non-integrable quantity is 7:3. The root mean square error of the prediction of the final model on the test set is 1.8MW, which improves the prediction accuracy by 65% compared with the traditional method.

[0031] In the embodiments of the present application, by performing data cleaning and normalization processing on the original frequency modulation instruction sequence, missing values, outliers, and noise interference in the data are effectively removed, ensuring the reliability of data quality. By decomposing the standardized frequency modulation sequence through a smoothing function, the complex non-linear sequence is decomposed into two parts: an approximately integrable quantity and a non-integrable quantity, which is convenient for separate processing and improves the prediction accuracy. The approximately integrable quantity is subjected to trigonometric function periodic conversion processing to convert the irregular time series into a combination of periodic functions with clear mathematical characteristics, enhancing the predictability of the signal. Through linear combination transformation processing of the periodic feature sequence and the non-integrable quantity using weight coefficients, the organic combination of periodic changes and random fluctuations is achieved, enabling the prediction model to capture both the regular and random characteristics of the signal simultaneously. A deep neural network is used to train the combined feature sequence, making full use of the advantages of the convolutional neural network to extract local features and the long short-term memory network to capture long-term dependence relationships, and a model with powerful prediction ability is constructed. In the real-time prediction stage, through preprocessing and inverse normalization processing, the accuracy and practicality of the prediction results are ensured. The overall solution significantly improves the prediction accuracy of the frequency modulation instruction through the system design of data preprocessing, feature extraction, model training, and prediction output. Experimental verification shows that the prediction RMSE of this method for frequency modulation instruction 1 and frequency modulation instruction 2 are 1.2 and 3.3 respectively, showing a significant improvement compared with 9.9 and 8.7 of the traditional model, providing a reliable decision-making basis for the efficient operation of the supercapacitor and lithium battery hybrid energy storage system. This improvement in prediction accuracy helps to optimize the power distribution strategy of the energy storage system, improve the frequency modulation performance of the system, extend the service life of the equipment, and reduce the operating cost.

[0032] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0033] (1) Perform missing value detection processing on the original frequency modulation instruction sequence to obtain the missing value position information, and perform linear interpolation processing on the missing value position information to obtain a complemented sequence;

[0034] (2) Perform outlier detection processing on the complemented sequence to obtain an outlier marking sequence, and perform median replacement processing on the outlier marking sequence to obtain a denoised sequence;

[0035] (3) Perform time window partitioning processing on the denoised sequence to obtain time series segments of a fixed length, and perform overlapping sliding processing on the time series segments of the fixed length to obtain sliding sequence samples;

[0036] (4) Perform amplitude statistical processing on the sliding sequence samples to obtain amplitude range parameters, and perform min-max normalization processing on the sliding sequence samples according to the amplitude range parameters to obtain a normalized sequence;

[0037] (5) Process the normalized sequence through Fourier transform to obtain frequency feature parameters, and perform band-pass filtering on the normalized sequence according to the frequency feature parameters to obtain a standardized frequency modulation sequence.

[0038] Specifically, perform missing value detection processing. By scanning the data points in the original frequency modulation instruction sequence, identify the positions where the values are empty, zero, or outside the physical significance range. The missing value detection uses a sliding time window scanning method. The window size is set to 10 seconds, and the data within each time window is checked for continuity. When it is found that the data at a certain time point is missing or abnormal, record the index position of that time point in the missing value position information table. After obtaining the missing value position information, use the linear interpolation method for data repair. The specific approach is to calculate the interpolation based on the valid data points before and after the missing point according to the proportional relationship of the time intervals. Linear interpolation ensures the continuity of the data and avoids sudden changes between the filled data and the original data. Perform outlier detection on the completed sequence using the box plot method to identify outliers. First, calculate the quartiles of the sequence, including the first quartile Q1, the median Q2, and the third quartile Q3, and calculate the interquartile range IQR = Q3 - Q1. Mark the data points less than (Q1 - 1.5×IQR) or greater than (Q3 + 1.5×IQR) as outliers to generate an outlier marking sequence. For the marked outliers, use the median replacement method for processing. The specific approach is to calculate the median of all normal data points within a 15-second time window before and after the outlier point and replace the outlier with this median. The median replacement method is insensitive to extreme values and can effectively maintain the overall distribution characteristics of the data.

[0039] After completing the outlier processing, divide the denoised sequence into time windows. Considering the dynamic characteristics of the frequency modulation instructions, set the time window length to 5 minutes. This length can capture both short-term fluctuation characteristics and not lose important trend information. When performing window division, use an overlapping sliding method, and set the overlap rate between adjacent windows to 75%, that is, the sliding step size is 75 seconds each time. Overlapping sliding can increase the correlation between samples and provide more training data. For each time window, extract the complete time series data segment within the window to form a time series segment of a fixed length. Through overlapping sliding processing, organize these time series segments into a sliding sequence sample set.

[0040] Perform amplitude statistical analysis on the sliding sequence samples to calculate the numerical distribution characteristics of the entire sample set. First, count the maximum value Vmax and the minimum value Vmin in the sample set, and calculate the numerical range span. At the same time, calculate the mean μ and the standard deviation σ. These statistics constitute the amplitude range parameters. Based on these parameters, perform min-max normalization on the sliding sequence samples to map all data into the interval [0, 1]. The mathematical expression of normalization is: for the original value x, the normalized value is equal to (x - Vmin) / (Vmax - Vmin). The normalization process eliminates the influence of dimensions and makes the data at different time periods comparable. Perform frequency domain analysis on the normalized sequence, and use the fast Fourier transform (FFT) method to convert the time domain signal to the frequency domain. Obtain the spectrum of the signal through FFT calculation, analyze the amplitude and phase information of each frequency component, and obtain the frequency characteristic parameters. The frequency characteristic parameters include the frequency values of the main frequency components, the corresponding amplitude magnitudes, and the phase angles. Design a band-pass filter according to the frequency characteristic parameters. The passband range of the filter covers the main frequency components of the signal, and at the same time filters out high-frequency noise and extremely low-frequency drift. The signal after filtering is the standardized frequency modulation sequence.

[0041] Taking the actual frequency modulation data of a certain thermal power plant as an example, within a one-hour sampling period, frequency modulation command data is collected at 1-second intervals, and a total of 3600 data points are obtained. Through missing value detection, it is found that there are 15 missing values, located at time points such as 123 seconds and 256 seconds. Perform linear interpolation on these positions. For example, for the missing value at 123 seconds, based on the value of 48.5 MW at 122 seconds and the value of 49.2 MW at 124 seconds, the interpolated value at 123 seconds is 48.85 MW. The outlier detection analysis shows that Q1 of the original data is -35 MW, Q3 is 35 MW, IQR is 70 MW, and 25 outlier points are detected. For example, the -85 MW that appears at 789 seconds exceeds the normal range and is replaced with the median -42 MW within a 15-second window before and after this point. Divide the processed sequence into windows of 5 minutes, obtaining 12 basic time windows. Through sliding processing with a 75% overlap rate, 45 sample segments are extended. Perform statistical analysis on these samples, and obtain a maximum value of 50 MW, a minimum value of -50 MW, a mean close to 0 MW, and a standard deviation of 28 MW. Map the values to the interval [0, 1] through min-max normalization. FFT analysis finds that the main frequency components are concentrated between 0.1 Hz and 0.3 Hz. Accordingly, the passband range of the designed band-pass filter is from 0.08 Hz to 0.35 Hz, filtering out the slow-varying drift below 0.08 Hz and the high-frequency noise above 0.35 Hz. The processed standardized frequency modulation sequence exhibits good periodicity and stability.

[0042] In a specific embodiment, the process of performing step S102 may specifically include the following steps:

[0043] (1) Process the standardized frequency modulation sequence through a Gaussian kernel function to obtain a smoothed sequence, and perform window length adaptive adjustment processing on the smoothed sequence to obtain kernel function parameters;

[0044] (2) Perform convolution operation processing on the smoothed sequence according to the kernel function parameters to obtain an approximately integrable quantity, and perform difference calculation processing on the standardized frequency modulation sequence and the approximately integrable quantity to obtain a non-integrable quantity.

[0045] Specifically, use a Gaussian kernel function to smooth the standardized frequency modulation sequence. The basic form of the Gaussian kernel function is a bilateral exponential decay function, which is bell-shaped in the time domain. The reason for choosing the Gaussian kernel function as the smoothing tool is that it has good local properties and smoothing characteristics, and will not introduce phase distortion in the processing of signals. During the processing of the Gaussian kernel function, the processing of each data point considers the weighted influence of other points in its neighborhood, and the weighting coefficient decays in a Gaussian distribution as the distance from the center point increases, thereby obtaining a preliminary smoothed sequence. The window length adaptive adjustment of the smoothed sequence is to obtain the optimal kernel function parameters. The window length directly affects the smoothing degree. An overly long window will lead to over-smoothing and loss of effective fluctuation information, while an overly short window cannot effectively suppress noise. The adaptive adjustment process is based on the local characteristics of the signal and uses the cross-validation method to determine the optimal parameters by comparing the smoothing effects under different window lengths. For regions with large fluctuations, automatically reduce the window length to retain fast-changing features; for regions with gentle changes, appropriately increase the window length to enhance the smoothing effect.

[0046] After determining the kernel function parameters, perform convolution operation processing on the smoothed sequence. The essence of the convolution operation is to perform sliding weighted averaging on the signal with the Gaussian kernel function, and the kernel function parameters determine the weighting method and range. Through the convolution operation, the high-frequency components of the signal are attenuated, while the main trend is retained, forming an approximately integrable quantity. The approximately integrable quantity reflects the basic change trend of the frequency modulation command and has good mathematical properties, which is convenient for subsequent periodic processing. Finally, perform difference calculation on the standardized frequency modulation sequence and the approximately integrable quantity to obtain a non-integrable quantity. The non-integrable quantity contains the fast fluctuation components that are smoothed out in the original sequence. Although these components do not have good integral properties, they contain important local feature information. Through this decomposition method, the complex frequency modulation command sequence is divided into two parts that can be processed, laying a foundation for subsequent feature extraction and prediction modeling.

[0047] Taking specific data processing as an example: The standardized frequency modulation sequence at a certain moment contains 100 data points within 10 seconds, and the numerical range is between [-1, 1]. Through initial processing with a Gaussian kernel function, a kernel function with a standard deviation of 0.5 seconds is selected for smoothing to obtain a preliminary smoothed sequence. During the adaptive adjustment process, the data is divided into 5 sub-segments, each sub-segment being 2 seconds long, and the local fluctuation intensity is calculated separately. For the sub-segment with a fluctuation intensity of 0.8, the standard deviation of the kernel function is adjusted to 0.3 seconds; for the sub-segment with a fluctuation intensity of 0.2, the standard deviation of the kernel function is adjusted to 0.7 seconds. Convolution operation is performed using the adjusted parameters, and the obtained approximate integrable quantity exhibits obvious trend characteristics, with an amplitude approximately 80% of the original sequence. Subtracting the original sequence from the approximate integrable quantity, the amplitude of the non-integrable quantity obtained is approximately 20% of the original sequence, mainly manifested as high-frequency fluctuation components. The entire processing process maintains signal energy conservation, providing input data with reliable quality for subsequent periodic conversion.

[0048] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0049] (1) Perform basic period estimation processing on the approximate integrable quantity to obtain period parameters, and segment the approximate integrable quantity according to the period parameters to obtain a segmented sequence;

[0050] (2) Perform harmonic decomposition processing on the segmented sequence to obtain the fundamental wave and harmonic components, and perform sine and cosine function fitting processing on the fundamental wave and harmonic components to obtain fitting coefficients;

[0051] (3) Perform superposition calculation processing on the fitting coefficients to obtain a period characteristic sequence.

[0052] Specifically, the main periodic characteristics of the sequence are determined through the fundamental period estimation process, and the autocorrelation analysis method is used to calculate the correlation coefficients of the sequence at different time delays. When the first significant peak appears in the correlation coefficient, the corresponding time delay is the fundamental period. According to the obtained period parameter, the approximate integrable quantity is segmented into equal-length segments, and the length of each segment is equal to the fundamental period, ensuring that the segmentation points are located at the similar phases of the waveform, so as to obtain multiple segmented sequences with complete periods. The harmonic decomposition of the segmented sequences is to extract the contributions of different frequency components. The fundamental wave component corresponds to the fundamental period of the sequence, and its frequency is 1 / T (T is the fundamental period). The harmonic components are integer multiples of the fundamental wave frequency, and usually the 5th harmonic is considered. Through the Fourier analysis method, each segmented sequence is decomposed into the fundamental wave and multiple harmonic components. The sine and cosine functions are respectively fitted to these components to solve the amplitude and phase parameters. The least squares method is used in the fitting process to obtain the corresponding sine term and cosine term coefficients for each frequency component, and these coefficients together form the fitting coefficient set. Finally, the fitting results of all frequency components are superimposed to realize the reconstruction of the periodic feature sequence. When performing the superimposition calculation, the importance of each component is considered. Usually, the weight of the fundamental wave component is the largest, and the weight of the high-order harmonics decreases as the frequency increases. The reconstructed periodic feature sequence retains the main periodic change characteristics of the original sequence, providing a reliable basis for subsequent prediction analysis.

[0053] Taking the actual frequency modulation data processing as an example: perform autocorrelation analysis on a certain approximate integrable quantity sequence. By calculating the correlation coefficients at different time delays (0 - 600 seconds), the first significant peak is found at 180 seconds, the correlation coefficient is 0.85, and the fundamental period is determined to be 180 seconds. According to this period parameter, the sequence is divided into multiple segments, and each segment contains 180 data points (sampling frequency 1Hz). Perform harmonic analysis on a single segmented sequence to obtain the fundamental wave frequency of 0.0056Hz, and the 2nd - 5th harmonic frequencies are 0.0112Hz, 0.0168Hz, 0.0224Hz, 0.0280Hz respectively. The amplitude of the fundamental wave component is 0.6, and the phase angle is 45 degrees. The amplitudes of the second to fifth harmonics are 0.3, 0.15, 0.08, 0.04 respectively, and the phase angles are 30 degrees, 60 degrees, 90 degrees, 120 degrees respectively. These components are superimposed according to the obtained amplitudes and phases to reconstruct the periodic feature sequence. The correlation coefficient between this sequence and the original data reaches 0.92, indicating a good reconstruction effect.

[0054] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0055] (1) Perform importance evaluation processing on the periodic feature sequence to obtain the periodic feature weight, and perform volatility analysis processing on the non-integrable quantity to obtain the volatility feature weight;

[0056] (2) Perform weighted processing on the periodic feature sequence according to the periodic feature weight to obtain a weighted periodic sequence, and perform weighted processing on the non-integrable quantity according to the fluctuation feature weight to obtain a weighted non-integrable sequence;

[0057] (3) Perform linear combination processing on the weighted periodic sequence and the weighted non-integrable sequence to obtain a combined feature sequence.

[0058] Specifically, weight assignment and combination of the periodic feature sequence and the non-integrable quantity are important links in accurate prediction. The weight of the periodic feature sequence is determined through importance assessment, and the assessment method is based on the degree of explanation of the periodic feature sequence for the changes in the original sequence. Specifically, calculate the cross-correlation coefficient between the periodic feature sequence and the original standardized frequency modulation sequence, analyze the energy distribution of the periodic feature sequence, and evaluate its prediction ability on different time scales. At the same time, perform volatility analysis on the non-integrable quantity, calculate its statistical features such as root mean square value, kurtosis, and skewness, and evaluate the intensity and distribution characteristics of random fluctuations, thereby obtaining the fluctuation feature weight. Perform weighted processing on the periodic feature sequence and the non-integrable quantity respectively according to the evaluated weight coefficients. For the periodic feature sequence, apply the periodic feature weight to each data point of the sequence to obtain a weighted periodic sequence. This process highlights the parts with significant periodicity in the sequence and enhances the expression of regular features. For the non-integrable quantity, apply the fluctuation feature weight to each fluctuation component in the sequence to obtain a weighted non-integrable sequence. This weighted process retains the meaningful random fluctuation information in the sequence and suppresses irrelevant noise interference.

[0059] Finally, linearly combine the weighted periodic sequence and the weighted non-integrable sequence, and use the normalized weight coefficients to ensure that the combined sequence maintains an appropriate numerical range. The linear combination process realizes the optimal fusion of periodic characteristics and random fluctuation characteristics through fine adjustment of the weights. The obtained combined feature sequence not only retains the main change rules of the frequency modulation instruction but also contains the necessary fluctuation information.

[0060] Taking specific data processing as an example: For the periodic feature sequence of a certain section of FM data, the importance is evaluated. By calculating the cross-correlation coefficient with the original sequence, a correlation of 0.85 is obtained, and the energy proportion reaches 75%. Based on this, the weight of the periodic feature is determined to be 0.7. The volatility analysis of the non-integrable quantity shows that its root mean square value is 0.3, kurtosis is 3.2, and skewness is 0.1, indicating that the fluctuation distribution is relatively uniform. The weight of the fluctuation feature is determined to be 0.3. The periodic feature sequence is weighted with a weight coefficient of 0.7, and the part of the original sequence with an amplitude of ±0.8 is adjusted to ±0.56. The non-integrable quantity is weighted with a weight coefficient of 0.3, and the part of the original fluctuation amplitude of ±0.4 is adjusted to ±0.12. Finally, the two weighted sequences are linearly combined, and the resulting combined feature sequence superimposes moderate random fluctuations within the periodic change interval. The overall change range of the sequence is controlled between ±0.7, which not only retains the main features of the original signal but also has good prediction value.

[0061] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0062] (1) Perform training set partitioning on the combined feature sequence to obtain a training sample set, and perform batch size setting on the training sample set to obtain data batches;

[0063] (2) Process the data batches through a three-layer convolutional neural network to obtain a feature mapping sequence, and process the feature mapping sequence through a bidirectional long short-term memory network to obtain time series features;

[0064] (3) Process the time series features through a fully connected layer to obtain an FM command prediction model.

[0065] Specifically, in the frequency modulation command prediction method for supercapacitor-coupled lithium battery energy storage, the deep neural network training of the combined feature sequence is the core link in constructing the prediction model. First, the combined feature sequence is divided into training sets in chronological order. The sliding window method is adopted, with the input window length set to 30 minutes and the prediction window length set to 5 minutes. The training sample set is constructed in a proportion of 70% for training, 15% for validation, and 15% for testing. The batch size is set for the training sample set. Considering the video memory limit and training efficiency, the batch size is set to 64, and the training samples are organized into corresponding data batches through a data loader. The constructed three-layer convolutional neural network is used to extract local features. The first layer of convolution uses 32 convolutional kernels of size 3×1 with a stride of 1 to extract features from the input sequence; the second layer of convolution uses 64 convolutional kernels of size 3×1 with a stride of 1 to increase the richness of features; the third layer of convolution uses 128 convolutional kernels of size 3×1 with a stride of 1 to further extract deep features. After each layer of convolution, a batch normalization layer and a ReLU activation function are connected, and a max pooling layer is used to reduce the data dimension. The feature map sequence output by the convolutional network contains the local temporal features of the frequency modulation command.

[0066] The bidirectional long short-term memory network is used to process the feature map sequence, which contains two layers of LSTM structures, and each layer contains 256 hidden units. The bidirectional structure can consider the forward and backward dependencies of the sequence simultaneously, improving the ability to extract temporal features. The LSTM unit contains an input gate, a forget gate, and an output gate, which can effectively handle the long-term dependence problem. Through the processing of the bidirectional LSTM, a feature vector encoding the long and short-term temporal relationships is obtained. Finally, a three-layer fully connected network is used for feature mapping. The first layer contains 512 neurons, the second layer contains 256 neurons, and the number of neurons in the third layer matches the dimension of the prediction target. A Dropout layer is added between the fully connected layers, with the dropout rate set to 0.3 to prevent overfitting. The linear activation function is used in the last layer to output the prediction result.

[0067] Taking the actual training process as an example: The frequency modulation data of a thermal power plant in a week contains 10,080 sampling points (sampling interval is 1 minute), and 8,640 training samples are obtained through the sliding window. After setting the batch size to 64, 135 training batches are formed. After the first layer of convolution processing, the size of the feature map is 28×32; after the second layer of convolution, it becomes 26×64; the third layer of convolution outputs a feature map sequence of 24×128. After the bidirectional LSTM processing, a 512-dimensional temporal feature vector is obtained. After mapping through the fully connected layer, a prediction sequence with a 5-minute resolution is finally output. Evaluated on the test set, the root mean square error of the model's prediction of the frequency modulation command within the next 5 minutes is 1.2 MW.

[0068] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0069] (1) Perform data preprocessing on the real-time frequency modulation command data to obtain a real-time standardized sequence, and perform forward calculation processing on the real-time standardized sequence through a frequency modulation command prediction model to obtain an original prediction value;

[0070] (2) Perform constraint check processing on the original prediction value to obtain an effective prediction result, and perform inverse normalization transformation on the effective prediction result to obtain a frequency modulation command prediction value.

[0071] Specifically, the processing of real-time frequency modulation command data is the last link to achieve accurate prediction. First, it is necessary to perform data preprocessing on the real-time frequency modulation command data, and this process is consistent with the data processing in the training stage, including data cleaning, normalization, and feature extraction. Detect missing values and handle outliers in the received real-time data, and perform min-max normalization using the same parameters as in the training stage to ensure the unity of data format and numerical range, obtaining a real-time standardized sequence. Input the standardized sequence into the frequency modulation command prediction model for forward calculation. The model sequentially passes through the convolutional layer, recurrent layer, and fully connected layer according to the feature mapping relationship established during training, and calculates to obtain the original prediction value. Performing constraint check on the original prediction value output by the model is an important step to ensure the effectiveness of the prediction result. Constraint check includes two aspects: physical constraint and logical constraint. Physical constraint ensures that the prediction value does not exceed the actual adjustment ability range of the system. For example, check whether the prediction value is within the maximum adjustment rate limit of the thermal power unit; logical constraint ensures the continuity and smoothness of the prediction value, avoiding unreasonable jumps. Perform inverse normalization transformation on the prediction result that passes the constraint check, map the numerical value in the standardized space back to the original value range, and obtain the final frequency modulation command prediction value.

[0072] Taking the actual prediction process as an example: The real-time frequency modulation command data received at a certain moment is preprocessed, and the original value range between -45 MW and +52 MW is normalized to the 0-1 interval. After forward calculation by the prediction model, an original prediction value of 0.7 is obtained. Through physical constraint check, it is found that the change rate corresponding to this value is 2.5 MW / minute, which does not exceed the 3 MW / minute adjustment limit of the unit. At the same time, through logical constraint check, it is confirmed that the change amplitude of the prediction value is within a reasonable range compared with the previous moment. Finally, through inverse normalization transformation, 0.7 is mapped back to the original dimension, obtaining a frequency modulation command prediction value of 35 MW. This prediction value not only satisfies the physical constraints of the system but also has good temporal continuity.

[0073] The above describes the frequency modulation command prediction method for supercapacitor-coupled lithium battery energy storage in the embodiments of the present application. Next, the frequency modulation command prediction system for supercapacitor-coupled lithium battery energy storage in the embodiments of the present application will be described. Please refer toFigure 2 , in an embodiment of the present application, an embodiment of the frequency modulation command prediction system for supercapacitor-coupled lithium battery energy storage includes:

[0074] A processing module 201, configured to perform data cleaning and normalization processing on the original frequency modulation command sequence to obtain a standardized frequency modulation sequence;

[0075] A decomposition module 202, configured to decompose the standardized frequency modulation sequence through a smoothing function to obtain an approximately integrable quantity and a non-integrable quantity;

[0076] A conversion module 203, configured to perform trigonometric periodic conversion processing on the approximately integrable quantity to obtain a periodic feature sequence, where the periodic feature sequence is composed of sine and cosine functions of multiple different frequencies;

[0077] A transformation module 204, configured to perform linear combination transformation processing on the periodic feature sequence and the non-integrable quantity through weight coefficients to obtain a combined feature sequence;

[0078] A training module 205, configured to perform training processing on the combined feature sequence through a deep neural network to obtain a frequency modulation command prediction model, where the frequency modulation command prediction model includes a three-layer convolutional neural network, a bidirectional long short-term memory network, and a fully connected layer;

[0079] A prediction module 206, configured to perform prediction processing on real-time frequency modulation command data through the frequency modulation command prediction model to obtain a prediction result, and perform anti-normalization processing on the prediction result to obtain a frequency modulation command prediction value.

[0080] Through the collaborative cooperation of the above-mentioned various components, by performing data cleaning and normalization on the original frequency modulation instruction sequence, missing values, outliers, and noise interference in the data are effectively removed, ensuring the reliability of data quality. Through the decomposition of the standardized frequency modulation sequence by a smoothing function, the complex non-linear sequence is decomposed into two parts: an approximately integrable quantity and a non-integrable quantity, which is convenient for separate processing and improves the prediction accuracy. The approximately integrable quantity is subjected to trigonometric function periodic conversion processing, converting the irregular time series into a combination of periodic functions with clear mathematical characteristics, enhancing the predictability of the signal. Through the linear combination transformation of the periodic feature sequence and the non-integrable quantity by the weight coefficient, the organic combination of periodic changes and random fluctuations is realized, enabling the prediction model to capture the regular and random characteristics of the signal simultaneously. The combined feature sequence is trained using a deep neural network, making full use of the advantages of the convolutional neural network to extract local features and the long short-term memory network to capture long-term dependence relationships, and a model with powerful prediction ability is constructed. In the real-time prediction stage, through preprocessing and anti-normalization processing, the accuracy and practicality of the prediction results are ensured. The overall solution significantly improves the prediction accuracy of the frequency modulation instruction through the system design of data preprocessing, feature extraction, model training, and prediction output. Experimental verification shows that the prediction RMSE of this method for frequency modulation instruction 1 and frequency modulation instruction 2 are 1.2 and 3.3 respectively, showing a significant improvement compared with 9.9 and 8.7 of the traditional model, providing a reliable decision-making basis for the efficient operation of the supercapacitor and lithium battery hybrid energy storage system. This improvement in prediction accuracy helps to optimize the power distribution strategy of the energy storage system, improve the frequency modulation performance of the system, extend the service life of the equipment, and reduce the operating cost.

[0081] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A frequency modulation instruction prediction method for supercapacitor coupled lithium battery energy storage, characterized in that: The frequency modulation instruction prediction method for supercapacitor coupled lithium battery energy storage includes: Perform data cleaning and normalization on the original frequency modulation instruction sequence to obtain a standardized frequency modulation sequence; Decomposing the standardized frequency modulation sequence by a smoothing function to obtain an approximate integrable quantity and an integrable quantity; The approximate integrable quantity is subjected to periodic conversion processing by trigonometric functions to obtain a periodic characteristic sequence, wherein the periodic characteristic sequence is composed of a plurality of sine and cosine functions of different frequencies; Performing linear merging transformation processing on the periodic characteristic sequence and the non-integrable quantity through weight coefficients to obtain a combined characteristic sequence; The combined feature sequence is trained through a deep neural network to obtain a frequency modulation instruction prediction model, wherein the frequency modulation instruction prediction model includes a three-layer convolutional neural network, a bidirectional long short-term memory network and a fully connected layer; The real-time frequency modulation instruction data is predicted by the frequency modulation instruction prediction model to obtain a prediction result, and the prediction result is denormalized to obtain a frequency modulation instruction prediction value; The real-time frequency modulation instruction data is predicted by the frequency modulation instruction prediction model to obtain a prediction result, and the prediction result is denormalized to obtain a frequency modulation instruction prediction value, including: Performing data preprocessing on the real-time frequency modulation instruction data to obtain a real-time standardized sequence, and performing forward calculation processing on the real-time standardized sequence through the frequency modulation instruction prediction model to obtain an original prediction value; The original prediction value is subjected to constraint checking to obtain a valid prediction result, and the valid prediction result is subjected to inverse normalization transformation to obtain the frequency modulation instruction prediction value.

2. The frequency modulation instruction prediction method for supercapacitor coupled lithium battery energy storage according to claim 1, characterized in that: The data cleaning and normalization processing of the original frequency modulation instruction sequence to obtain a standardized frequency modulation sequence includes: Performing missing value detection processing on the original frequency modulation instruction sequence to obtain missing value position information, and performing linear interpolation processing on the missing value position information to obtain a completed sequence; Performing outlier detection processing on the completed sequence to obtain an outlier marker sequence, and performing median replacement processing on the outlier marker sequence to obtain a denoised sequence; Performing time window division processing on the denoised sequence to obtain time sequence segments of fixed length, and performing overlapping sliding processing on the time sequence segments of fixed length to obtain sliding sequence samples; Performing amplitude statistical processing on the sliding sequence samples to obtain an amplitude range parameter, and performing min-max normalization processing on the sliding sequence samples according to the amplitude range parameter to obtain a normalized sequence; The normalized sequence is processed by Fourier transform to obtain frequency characteristic parameters, and the normalized sequence is subjected to bandpass filtering according to the frequency characteristic parameters to obtain the standardized frequency modulation sequence.

3. The frequency modulation instruction prediction method for supercapacitor coupled lithium battery energy storage according to claim 1, characterized in that: Decomposing the standardized frequency modulation sequence by a smoothing function to obtain approximate integrable quantities and integrable quantities includes: Processing the standardized frequency modulation sequence by a Gaussian kernel function to obtain a smoothed sequence, and processing the smoothed sequence by adaptively adjusting the window length to obtain kernel function parameters; The smoothed sequence is subjected to convolution operation according to the kernel function parameters to obtain an approximate integrable quantity, and the standardized frequency modulation sequence and the approximate integrable quantity are subjected to difference calculation to obtain an integrable quantity.

4. The frequency modulation instruction prediction method for supercapacitor coupled lithium battery energy storage according to claim 1, characterized in that: The approximate integrable quantity is subjected to periodic conversion processing by a trigonometric function to obtain a periodic characteristic sequence, wherein the periodic characteristic sequence is composed of a plurality of sine and cosine functions of different frequencies, including: Performing basic period estimation processing on the approximate integrable quantity to obtain period parameters, and performing segmentation processing on the approximate integrable quantity according to the period parameters to obtain a segmented sequence; Performing harmonic decomposition processing on the segmented sequence to obtain fundamental wave and harmonic wave components, and performing sine-cosine function fitting processing on the fundamental wave and harmonic wave components to obtain fitting coefficients; The fitting coefficients are subjected to superposition calculation processing to obtain the periodic characteristic sequence.

5. The frequency modulation instruction prediction method for supercapacitor coupled lithium battery energy storage according to claim 1, characterized in that: The linear merging and transformation processing of the periodic feature sequence and the non-integrable quantity by weight coefficients to obtain a combined feature sequence includes: Performing importance assessment processing on the periodic characteristic sequence to obtain a periodic characteristic weight, and performing volatility analysis processing on the non-integrable quantity to obtain a volatility characteristic weight; Performing weighted processing on the periodic characteristic sequence according to the periodic characteristic weight to obtain a weighted periodic sequence, and performing weighted processing on the integrable quantity according to the fluctuation characteristic weight to obtain a weighted integrable sequence; The weighted periodic sequence and the weighted non-integrable sequence are linearly combined to obtain the combined characteristic sequence.

6. The frequency modulation instruction prediction method for supercapacitor coupled lithium battery energy storage according to claim 1, characterized in that: The combined feature sequence is trained through a deep neural network to obtain a frequency modulation instruction prediction model, wherein the frequency modulation instruction prediction model includes a three-layer convolutional neural network, a bidirectional long short-term memory network and a fully connected layer, including: Performing training set division processing on the combined feature sequence to obtain a training sample set, and performing batch size setting processing on the training sample set to obtain a data batch; Processing the data batches through a three-layer convolutional neural network to obtain a feature mapping sequence, and processing the feature mapping sequence through a bidirectional long short-term memory network to obtain a time series feature; The timing features are processed through a fully connected layer to obtain the frequency modulation instruction prediction model.

7. A frequency modulation instruction prediction system for supercapacitor coupled lithium battery energy storage, used to implement the frequency modulation instruction prediction method for supercapacitor coupled lithium battery energy storage as described in any one of claims 1-6, characterized in that: The frequency modulation instruction prediction system for supercapacitor coupled lithium battery energy storage includes: A processing module is used to perform data cleaning and normalization processing on the original frequency modulation instruction sequence to obtain a standardized frequency modulation sequence; A decomposition module, used for decomposing the standardized frequency modulation sequence through a smoothing function to obtain an approximate integrable quantity and an integrable quantity; A conversion module, used for performing periodic conversion processing on the approximate integrable quantity through trigonometric functions to obtain a periodic characteristic sequence, wherein the periodic characteristic sequence is composed of a plurality of sine and cosine functions of different frequencies; A transformation module, used for performing a linear merging transformation process on the periodic feature sequence and the non-integrable quantity through a weight coefficient to obtain a combined feature sequence; A training module, used for training the combined feature sequence through a deep neural network to obtain a frequency modulation instruction prediction model, wherein the frequency modulation instruction prediction model includes a three-layer convolutional neural network, a bidirectional long short-term memory network and a fully connected layer; A prediction module, used to perform prediction processing on the real-time frequency modulation instruction data through the frequency modulation instruction prediction model to obtain a prediction result, and perform denormalization processing on the prediction result to obtain a frequency modulation instruction prediction value; The prediction module is used to perform data preprocessing on the real-time frequency modulation instruction data to obtain a real-time standardized sequence, and to perform forward calculation processing on the real-time standardized sequence through the frequency modulation instruction prediction model to obtain an original prediction value; to perform constraint checking processing on the original prediction value to obtain a valid prediction result, and to perform inverse normalization transformation processing on the valid prediction result to obtain the frequency modulation instruction prediction value.

Images