Energy storage frequency modulation instruction prediction system

Abstract

The application discloses a kind of energy storage frequency modulation instruction prediction systems, comprising: frequency modulation instruction acquisition module, for obtaining frequency modulation instruction sequence;Neural network improvement module is used to improve neural network according to the fluctuation of the frequency modulation instruction sequence;Prediction module is used to predict the frequency modulation instruction sequence by improved neural network, and the prediction result is applied to the control of energy storage system.The energy storage frequency modulation instruction prediction system of the embodiment of the present application, by the fluctuation of frequency modulation instruction sequence, improves the neural network for prediction, so that it can adaptively adjust the fusion weight of history memory and current input, to improve the accuracy of prediction.

Description

Technical Field

[0001] This disclosure belongs to the field of energy storage frequency regulation prediction technology, specifically relating to an energy storage frequency regulation command prediction system. Background Technology

[0002] Molten salt energy storage frequency regulation command prediction refers to establishing a predictive model by analyzing historical frequency regulation commands and related data to predict future frequency regulation command demands, thereby optimizing the control strategy of the hybrid energy storage system and improving its frequency regulation performance and efficiency. Its advantages mainly include the following: First, it can improve system performance and achieve rapid response, meaning it can quickly adjust output power in a short time to respond rapidly to changes in grid frequency, effectively balancing supply and demand differences in the power system and maintaining grid frequency stability; it can improve regulation accuracy, enabling more precise tracking of frequency regulation commands, reducing regulation errors, and improving the power supply quality and stability of the grid; it can enhance system stability, by rationally allocating the power and energy of different energy storage devices, reducing the overall risk of the system, improving system reliability and stability, and reducing the risk of system failure due to the failure or performance degradation of a single energy storage device. Second, it can extend equipment lifespan and optimize charging and discharging strategies. Based on the prediction results, reasonable charging and discharging strategies can be formulated to avoid overcharging and discharging or frequent charging and discharging of energy storage devices, thereby extending their service life; it can balance equipment usage, rationally allocating the usage frequency and load of different energy storage devices, making the aging of each device relatively even, reducing equipment maintenance costs and replacement frequency. Finally, it can also reduce costs and improve economic efficiency. By improving frequency regulation performance and efficiency, hybrid energy storage systems can obtain more frequency regulation benefits, while reducing power outage losses and equipment damage costs caused by grid frequency instability. It can also optimize investment costs. Based on the forecast results, the capacity and quantity of different energy storage devices can be reasonably configured to avoid over-investment and waste of resources, thereby improving the return on investment.

[0003] Traditional prediction methods for FM sequences often employ VMD decomposition or CEEMD decomposition to decompose the original FM signal into a series of subsequences IMF1, IMF2, IMF3, ..., IMF. K The data is then fed into a neural network for prediction, and the final predictions are then superimposed to obtain the final result. However, existing data decomposition techniques (VMD, CEEMD) have the following inherent drawbacks: real-time processing limitations, neither of which is suitable for real-time applications due to their long computation time; challenges in complex signal processing, as they may not be effective in decomposing signals with high-frequency noise, sudden fluctuations, or high nonlinearity; and experience dependence, as parameter adjustments rely on user experience and have limited automation. Summary of the Invention

[0004] This disclosure aims to at least address one of the technical problems existing in the prior art, and to provide a method and system for predicting energy storage frequency regulation commands.

[0005] This disclosure provides a method for predicting frequency regulation commands for energy storage, the method comprising: Obtain the frequency modulation command sequence; The neural network is improved based on the fluctuation of the frequency modulation command sequence; The frequency modulation command sequence is predicted by an improved neural network, and the prediction results are applied to the control of the energy storage system.

[0006] Optionally, the improvement of the neural network based on the variability of the frequency modulation command sequence includes: The discrete frequency modulation command sequence is coupled into a continuous global function through a dynamic piecewise frequency modulation fitting method; The fluctuation degree is obtained according to the global function; The GRU network is improved based on the aforementioned fluctuation.

[0007] Optionally, coupling the discrete frequency modulation command sequence into a continuous global function using a dynamic piecewise frequency modulation fitting method includes: The total duration of the frequency modulation command sequence is divided into multiple windows; The number of extreme points within each window is obtained by scanning the extreme points within that window. Calculate the frequency of extreme points occurring within the window based on the number of extreme points within the window; Calculate the change in the frequency of extreme points occurring within the window; The sliding basis function is obtained based on the change in frequency; The global function is obtained by concatenating the basis functions of each window in the sliding basis function.

[0008] Optionally, the frequency modulation command sequence is P=[X1,X2,X3,.X...]. i ..,X N (N>1000), where the subscript indicates the number of samples and the sampling interval is 1 second; The step of dividing the total duration of the frequency modulation command sequence into multiple windows includes: dividing the total duration into K windows, where K=N / W, and W is the window width that is divisible by N; The step of calculating the frequency of extreme points occurring within the window based on the number of extreme points within the window includes: , in, This is the starting frequency of the nth window, which is the frequency at which extreme points occur within the nth window.

[0009] Optionally, calculating the change in the frequency of extreme points occurring within the window includes: , in, Let be the rate of change of frequency within the nth window.

[0010] Optionally, obtaining the sliding basis function based on the frequency change includes: , in, Let n be the basis function for the nth window. For amplitude, Let n be the start time of the nth window. T represents the total time span.

[0011] Optionally, concatenating the basis functions of each window in the sliding basis functions to obtain the global function includes concatenating the basis functions of each window into a global function according to the principle of phase continuity: , , Here, F(t) is a global function.

[0012] Optionally, obtaining the fluctuation degree according to the global function includes: , Where Q represents the variability, To adapt to the Gaussian window width, for tanh is the hyperbolic tangent function. The tunneling threshold, Let F(t) be the first derivative at t. Let F(t) be the second derivative at t. is a learnable scalar gain coefficient.

[0013] Optionally, improving the GRU network based on the fluctuation degree includes: , in, The update gate output vector at time step t. The Sigmoid activation function is used. For trainable weight matrix, The sampled value of the frequency modulation command sequence at time t. Let be the hidden state vector at time t-1. Hidden state vector To update the weight matrix of the gate, Let be the frequency modulation command sequence at time t. For the newly added trainable weight matrix, This is element-wise multiplication.

[0014] Another aspect of this disclosure provides an energy storage frequency regulation command prediction system, comprising: The frequency modulation command acquisition module is used to acquire the frequency modulation command sequence; A neural network improvement module is used to improve the neural network based on the variability of the frequency modulation command sequence; The prediction module is used to predict the frequency modulation command sequence using an improved neural network and apply the prediction results to the control of the energy storage system.

[0015] In a method and system for predicting energy storage frequency modulation commands according to an embodiment of this disclosure, the predictive neural network is improved by the fluctuation of the frequency modulation command sequence, so that it can adaptively adjust the fusion weight of historical memory and current input, thereby improving the accuracy of prediction. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of an energy storage frequency regulation command prediction method according to an embodiment of the present disclosure; Figure 2 This is a schematic diagram of the structure of an energy storage frequency regulation command prediction system according to another embodiment of the present disclosure. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions of this disclosure, the disclosure will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0018] Unless otherwise specifically stated, the technical or scientific terms used in this disclosure should be understood in their ordinary sense by one of ordinary skill in the art to which this disclosure pertains. The terms "comprising" or "including," as used in this disclosure, do not limit the shapes, numbers, steps, actions, operations, components, elements, and / or groups thereof mentioned, nor do they exclude the appearance or inclusion of one or more other different shapes, numbers, steps, actions, operations, components, elements, and / or groups thereof, or the inclusion of these. In the description of this disclosure, "a plurality of" means two or more, unless otherwise expressly specified.

[0019] In some descriptions of the invention, unless otherwise expressly specified and limited, terms such as “installation,” “connection,” “linking,” or “fixing” are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect through an intermediate medium, which can be the internal connection of two elements or the interaction between two elements.

[0020] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this disclosure. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale, and techniques, methods, and devices known to those skilled in the art may not be discussed in detail; however, where appropriate, the illustrated techniques, methods, and devices should be considered part of the specification. In all the examples shown and discussed herein, any other specific example may have different values. It should be noted that similar symbols and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0021] Before discussing in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when an operation is completed, but it may also have additional steps not included in the figures. The process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0022] Below, we will refer to Figure 1 This disclosure describes a method for predicting energy storage frequency regulation commands according to an embodiment of the present disclosure.

[0023] For example, such as Figure 1 As shown, a method for predicting energy storage frequency regulation commands includes: S100: Obtain the frequency modulation command sequence.

[0024] For example, the frequency modulation command sequence is P=[X1,X2,X3,.X... i ..,X N (N>1000), where the subscripts 1,2,3…i…N of X represent the number of samples, and the sampling interval is 1 second.

[0025] S200: Improve the neural network based on the fluctuation of the frequency modulation command sequence.

[0026] For example, the neural network is a GRU network, which is a gated recurrent unit.

[0027] S300: The frequency modulation command sequence is predicted by an improved neural network, and the prediction results are applied to the control of the energy storage system.

[0028] Specifically, the frequency modulation command sequence P is input into the GRU network for prediction, that is, [X1, X2, X3, ..., X...] i ..,X N The data are fed into the improved GRU for prediction, and the final prediction results are obtained. The prediction results are then applied to the control of the energy storage system.

[0029] In a method for predicting energy storage frequency modulation commands according to an embodiment of this disclosure, the predictive neural network is improved by the fluctuation of the frequency modulation command sequence, so that it can adaptively adjust the fusion weight of historical memory and current input, thereby improving the accuracy of prediction.

[0030] The specific steps of a method for predicting frequency regulation commands for energy storage will be further described below.

[0031] For example, step S200 further includes: S210: The discrete frequency modulation command sequence P is coupled into a continuous global function through a dynamic segmented frequency modulation fitting method.

[0032] S220: Obtain the fluctuation degree according to the global function.

[0033] S230: Improve the GRU network based on the said undulation.

[0034] For example, step S210 further includes: S211: Divide the total duration of the frequency modulation command sequence into multiple windows.

[0035] Specifically, the total duration of the frequency instruction sequence P is divided into K windows, where K = N / W, and W is the window width that is divisible by N.

[0036] S212: Scan the extreme points in each window to obtain the number of extreme points in the window.

[0037] Specifically, the extreme points include maximum and minimum values. For example, the number of extreme points can be the sum of the number of maximum and minimum values, or it can be the number of maximum values ​​or the number of minimum values. Those skilled in the art can choose the specific setting method for the number of extreme points according to the actual use.

[0038] S213: Calculate the frequency of occurrence of extreme points within the window based on the number of extreme points within the window.

[0039] Specifically, the frequency of extreme points occurring within the window, i.e., the frequency parameter, is: , in, This is the starting frequency of the nth window, which is the frequency at which extreme points occur within the nth window.

[0040] It should be noted that in the process of estimating the frequency of extreme points, this disclosure directly calculates the frequency by the number of time-domain extremes, reducing the amount of computation by 90%, improving computational efficiency, and thus improving the real-time performance of the prediction.

[0041] S214: Calculate the change in the frequency of extreme points occurring within the window.

[0042] Specifically, the frequency of extreme points occurring within the window varies as follows: , in, Let be the rate of change of frequency within the nth window.

[0043] S215: Obtain the sliding basis function based on the change in frequency.

[0044] Specifically, a sliding basis function is constructed to generate a frequency-modulated waveform that slides over time, covering local features of the signal: , in, Let n be the basis function for the nth window. For amplitude, for example, it can be taken as 80% of the maximum absolute value of the data within the nth window. The starting frequency of the nth window. Let be the rate of change of frequency within the nth window. Let n be the start time of the nth window. W represents the window width. For example, the window width W can be 10 seconds, meaning 10 points are collected. T represents the total time span. For example, it can be set to collect one point per second, in which case T=N. t represents the current time. Since the sampling interval is 1 second, t has the same index as the first value of the window, which is [X1, X2, X3, ..., X...]. i ..,X N The subscripts are the same.

[0045] It should be noted that this disclosure uses dynamic frequency modulation basis functions and linear frequency modulation segments (Chirp) to cover local features, requiring only 2 parameters / window. In other words, only the starting frequency and the rate of frequency change are needed to express the signal, achieving the goal of capturing the essential features of the signal with very few parameters and reducing the amount of computation.

[0046] S216: Concatenate the basis functions of each window in the sliding basis function to obtain the global function.

[0047] Specifically, the basis functions of each window are concatenated into a global function according to the principle of phase continuity, making the function continuous and differentiable, in preparation for the next step.

[0048] First, the initial phase of the next window is automatically corrected according to the phase compensation rules: , Then, the basis functions of each window are concatenated into a global function according to the principle of phase continuity: , Here, F(t) is a global function.

[0049] In this step, phase self-synchronization is achieved through phase compensation, and seamless splicing is achieved through endpoint phase matching, avoiding the phase jump of the traditional segmentation method.

[0050] In a method for predicting frequency modulation commands for energy storage according to an embodiment of this disclosure, the frequency is directly calculated by the number of time-domain extrema during the extreme point frequency estimation process, reducing the computational load by 90% and improving computational efficiency; a dynamic frequency modulation basis function is used, and local features are covered by linear frequency modulation segments (Chirp), requiring only 2 parameters / window, that is, only the starting frequency and the frequency change rate are needed to complete the expression of the signal, achieving the goal of capturing the essential features of the signal with very few parameters and reducing the computational load; phase self-synchronization is achieved through phase compensation, and seamless splicing is achieved through endpoint phase matching, avoiding the phase jump of the traditional segmented method.

[0051] The specific steps of a method for predicting frequency regulation commands for energy storage will be further described below.

[0052] For example, after coupling in step S210, the original frequency modulation sequence P=[X1,X2,X3,.X... i ..,X N The original sequence P is coupled into a differentiable F(t), which can now be used to evaluate the fluctuation Q of the original sequence P. Specifically, step S220 is as follows: , Where Q represents the variability; To adapt to the Gaussian window width; ; tanh is the hyperbolic tangent function, which maps the difference to the interval (0,1); The tunneling threshold is set as the Planck constant, which is a reduction of the tunneling threshold, based on the Heisenberg uncertainty principle. Let F(t) be the first derivative at point t; Let F(t) be the second derivative at point t; This is a learnable scalar gain coefficient; an example value could be 1.1.

[0053] For example, the adaptive Gaussian window width It is associated with local variance, specifically: , in, This is an empirical coefficient, for example, a typical value of 0.3 can be taken; Var is the calculated value. The variance.

[0054] In a frequency modulation command prediction method for energy storage according to an embodiment of this disclosure, the sensitivity is dynamically modulated by the second derivative to achieve immunity to slowly varying interference, reducing the false alarm rate of abrupt changes by 75% and enhancing curvature adaptation; combined with adaptive window width and quantum tunneling effect, the capture delay of step-type abrupt changes is shortened to 0.3 seconds, which is 4 times faster than the variance method, achieving microsecond-level response; it has physical interpretability, and the parameter design is related to the Heisenberg uncertainty principle, providing a cross-scale quantum mechanical interpretation framework for frequency modulation control.

[0055] The specific steps of a method for predicting frequency regulation commands for energy storage will be further described below.

[0056] For example, step S230 specifically includes dynamically modulating the update gate of the GRU with the fluctuation Q so that it can adaptively adjust the fusion weights of the historical memory and the current input.

[0057] The standard formula for the update gate of a standard GRU is as follows: , in, The update gate output vector at time step t. The Sigmoid activation function is used. For trainable weight matrix, The i-th sample value of the frequency modulation command sequence. Let be the hidden state vector at time t-1. Hidden state vector To update the weight matrix of the gate, To update the bias vector of the gate.

[0058] After obtaining the fluctuation degree, the update gate of the standard GRU is improved based on the fluctuation degree, resulting in the following formula: , in, This is a learnable scalar gain coefficient, with an example value of 1.1. Its function is to dynamically adjust the influence of fluctuation on the update gate, and it is automatically optimized through training. The update gate output vector for time step t; Use the Sigmoid activation function; The weight matrix is ​​trainable. The sampled value of the frequency modulation command sequence at time t; Let be the hidden state vector at time t-1; Hidden state vector Update the weight matrix of the gate; Let be the frequency modulation command sequence at time t; This is a newly added trainable weight matrix used to map the hidden state to the modulation space; This is element-wise multiplication, also known as the Hadamard product.

[0059] in, For standard GRU update gates, This is a newly added dynamic modulation term.

[0060] In a method for predicting energy storage frequency modulation commands according to an embodiment of this disclosure, the update gate of the GRU is dynamically modulated by the fluctuation degree Q, so that it can adaptively adjust the fusion weight of historical memory and current input.

[0061] To further verify the advantages of the energy storage frequency regulation command prediction method disclosed herein, the present disclosure uses the method disclosed herein and the GRU prediction method to predict the energy storage frequency regulation sequence, and the results are shown in Table 1 and Table 2 below.

[0062] The frequency modulation command sequences were acquired from a power plant in Inner Mongolia. Frequency modulation command sequence 1 was acquired from 4:00 AM to 6:00 PM on a certain day in November 2024, with one data point collected every second. Frequency modulation sequence 2 was acquired from 12:00 AM to 10:00 PM on a certain day in February 2025, with one data point collected every second.

[0063] Table 1 Comparison of Experimental Data

[0064] Table 2 Four evaluation indicators

[0065] N represents the sample size. and These represent the actual value and the predicted value at time n, respectively.

[0066] The experimental results show that all four evaluation indicators have decreased, indicating that the model proposed in this publication can significantly improve prediction accuracy.

[0067] The following will combine Figure 2 A frequency regulation command prediction system for energy storage is described according to another embodiment of the present disclosure.

[0068] For example, such as Figure 2 As shown, an energy storage frequency regulation command prediction system includes: Frequency modulation command acquisition module 100 is used to acquire frequency modulation command sequences; The neural network improvement module 200 is used to improve the neural network based on the undulation of the frequency modulation command sequence; The prediction module 300 is used to predict the frequency modulation command sequence through an improved neural network and apply the prediction results to the control of the energy storage system.

[0069] In an energy storage frequency modulation command prediction system according to an embodiment of this disclosure, the prediction neural network is improved by the fluctuation of the frequency modulation command sequence, so that it can adaptively adjust the fusion weight of historical memory and current input, thereby improving the prediction accuracy.

[0070] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0071] It is understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of this disclosure, and this disclosure is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of this disclosure, and these modifications and improvements are also considered to be within the scope of protection of this disclosure.

Claims

1. A frequency regulation command prediction system for energy storage, characterized in that, include: The frequency modulation command acquisition module is used to acquire the frequency modulation command sequence; A neural network improvement module is used to improve the neural network based on the variability of the frequency modulation command sequence; The prediction module is used to predict the frequency modulation command sequence using an improved neural network and apply the prediction results to the control of the energy storage system.

2. The energy storage frequency regulation command prediction system according to claim 1, characterized in that, The neural network improved based on the fluctuation of the frequency modulation command sequence includes: The discrete frequency modulation command sequence is coupled into a continuous global function through a dynamic piecewise frequency modulation fitting method; The fluctuation degree is obtained according to the global function; The GRU network is improved based on the aforementioned fluctuation.

3. The energy storage frequency regulation command prediction system according to claim 2, characterized in that, The step of coupling the discrete frequency modulation command sequence into a continuous global function through a dynamic piecewise frequency modulation fitting method includes: The total duration of the frequency modulation command sequence is divided into multiple windows; The number of extreme points within each window is obtained by scanning the extreme points within that window. Calculate the frequency of extreme points occurring within the window based on the number of extreme points within the window; Calculate the change in the frequency of extreme points occurring within the window; The sliding basis function is obtained based on the change in frequency; The global function is obtained by concatenating the basis functions of each window in the sliding basis function.

4. The energy storage frequency regulation command prediction system according to claim 3, characterized in that, The frequency modulation command sequence is P=[X1,X2,X3,.X...]. i ..,X N (N>1000), where the subscript indicates the number of samples and the sampling interval is 1 second; The step of dividing the total duration of the frequency modulation command sequence into multiple windows includes: dividing the total duration into K windows, where K=N / W, and W is the window width that is divisible by N; The step of calculating the frequency of extreme points occurring within the window based on the number of extreme points within the window includes: ， in, This is the starting frequency of the nth window, which is the frequency at which extreme points occur within the nth window.

5. The energy storage frequency regulation command prediction system according to claim 4, characterized in that, The calculation of the change in the frequency of extreme points occurring within the window includes: ， in, Let be the rate of change of frequency within the nth window.

6. The energy storage frequency regulation command prediction system according to claim 5, characterized in that, The step of obtaining the sliding basis function based on the frequency change includes: ， in, Let n be the basis function for the nth window. For amplitude, Let n be the start time of the nth window. T represents the total time span.

7. The energy storage frequency regulation command prediction system according to claim 6, characterized in that, The step of concatenating the basis functions of each window in the sliding basis functions to obtain the global function includes concatenating the basis functions of each window into the global function according to the principle of phase continuity: ， ， Here, F(t) is a global function.

8. The energy storage frequency regulation command prediction system according to claim 7, characterized in that, The step of obtaining the fluctuation degree according to the global function includes: ， Where Q represents the variability, To adapt to the Gaussian window width, for tanh is the hyperbolic tangent function. The tunneling threshold, Let F(t) be the first derivative at t. Let F(t) be the second derivative at t. is a learnable scalar gain coefficient.

9. The energy storage frequency regulation command prediction system according to claim 8, characterized in that, The improvement of the GRU network based on the fluctuation includes: ， in, The update gate output vector at time step t. For the Sigmoid activation function, For trainable weight matrix, The sampled value of the frequency modulation command sequence at time t. Let be the hidden state vector at time t-1. Hidden state vector To update the weight matrix of the gate, Let be the frequency modulation command sequence at time t. For the newly added trainable weight matrix, This is element-wise multiplication.

Images