Energy storage frequency modulation instruction prediction method and system based on chaotic phase injection

Abstract

The invention relates to the technical field of power system energy storage equipment, in particular to an energy storage frequency modulation instruction prediction method and system based on chaotic phase injection. The method comprises the steps of selecting a candidate period to segment an original frequency modulation instruction to generate a sub-sequence set, calculating the dynamic time warping distance of all extracted sequence pairs and calculating and obtaining the similarity integral of the candidate period, screening a real period through standardized saliency test, selecting an optimal period with the maximum saliency value, and obtaining the optimal frequency modulation instruction. And calculating a consistency index in combination with the permutation entropy of the optimal period sub-sequence set, reconstructing an original frequency modulation instruction based on chaotic phase injection, and inputting a reconstructed sequence into the pre-trained GRU network to output a predicted value. According to the method, the nonlinear time sequence alignment problem is solved through the dynamic time warping distance, the chaotic intensity and the periodic disturbance are controlled through the PCI, anti-noise periodic detection and chaotic feature decoupling are achieved, finally, the prediction precision is improved, and response lag is eliminated.

Description

Technical Field

[0001] The present invention relates to the technical field of power system energy storage equipment, and in particular to a method and system for predicting energy storage frequency modulation instructions based on chaotic phase injection. Background Art

[0002] The core purpose of energy storage frequency regulation command prediction is to resolve the three major contradictions of regulation capacity mismatch, equipment life reduction, and limited economic benefits in the traditional passive response mode. By proactively predicting the dynamic characteristics of the power grid frequency regulation command, it can achieve energy storage system SOC (State of Charge) pre-control, adaptive optimization of charging and discharging strategies, and precise tracking of regulation accuracy, thereby improving the reliability and economy of energy storage participation in power system frequency regulation.

[0003] The existing AI-based energy storage frequency modulation instruction prediction has the following problems: (1) Timing alignment deviation causes strategy oscillation: There is usually a dynamic time lag (typical value 3-5 seconds) between the frequency modulation instruction and the energy storage response, and the existing time series prediction algorithms such as TCN (Temporal Convolutional Network) and Seq2Seq (Sequence to Sequence) use a fixed sliding window mechanism and cannot achieve dynamic phase matching of subsequences. Moreover, simulation display and timing misalignment will also lead to frequent switching of charging and discharging strategies; (2) Insufficient decoupling of chaotic features reduces prediction accuracy: Frequency modulation instructions usually imply chaotic characteristics of multiple sources such as grid inertia and new energy penetration, and their maximum Lyapunov exponent (Lyapunov exponent) is >0.15, but traditional AI models only extract explicit correlations through the attention mechanism and fail to quantify the injected chaotic phase components; (3) Lack of multi-time scale coordination: Existing methods usually independently construct minute-level and second-level prediction models. The lack of cross-scale feature interaction mechanism increases the probability of conflict between SOC planning and real-time command response, which easily triggers capacity over-limit alarms. Summary of the Invention

[0004] The technical problem to be solved by the embodiments of the present invention is to provide a method and system for predicting energy storage frequency modulation instructions based on chaotic phase injection, so as to solve the problems of low prediction accuracy and delayed response caused by timing alignment deviation, insufficient decoupling of chaotic characteristics and lack of multi-time scale coordination in the existing technology.

[0005] The present invention discloses a method for predicting energy storage frequency modulation instructions based on chaotic phase injection, comprising: Selecting a group of candidate cycles, and using each of the candidate cycles to segment the collected original frequency modulation instructions, to obtain a set of subsequences corresponding to the candidate cycles; Combining subsequences in the subsequence set in pairs to obtain sequence pairs, calculating the dynamic time warping distance of each of the sequence pairs, and calculating the similarity integral of the candidate period based on the dynamic time warping distances of all the sequence pairs; Obtaining significance of the candidate periods according to the similarity integral calculation of the candidate periods, screening out real periods from all the candidate periods according to the significance of the candidate periods, and determining the real period with the maximum significance value as the optimal period; Calculating and obtaining the permutation entropy of the subsequence set corresponding to the optimal period, and calculating and obtaining the consistency index of the optimal period by combining the significance of the optimal period and the permutation entropy; Based on the consistency index of the optimal period, a chaotic phase injection method is used to reconstruct the original frequency modulation instruction to obtain a reconstructed sequence; The reconstructed sequence is input into a pre-trained GRU network, and the output is a predicted value of the original frequency modulation instruction at a preset prediction duration.

[0006] Optionally, the step of dividing the collected original frequency modulation instructions by each candidate cycle to obtain a set of subsequences corresponding to the candidate cycles includes: The number of divisions of the original frequency modulation instruction is determined according to the total number of sampling times of the original frequency modulation instruction and the candidate cycle. The function expression of the number of divisions of the original frequency modulation instruction is:

[0007] Where K represents the number of subsequences after segmentation, N represents the total number of sampling times of the original frequency modulation instruction, T represents the candidate cycle, Indicates rounding down; According to the determined number of divisions, the original frequency modulation instruction is divided into a corresponding number of subsequences using the candidate cycles, and the divided subsequences are arranged in a division order to form the subsequence set.

[0008] Optionally, calculating the dynamic time warping distance of each of the sequence pairs and obtaining the similarity integral of the candidate period based on the dynamic time warping distances of all the sequence pairs includes: For each of the sequences, a dynamic time warping distance is obtained by minimizing the cumulative path distance through dynamic programming; The average similarity is calculated based on the dynamic time warping distance of all the sequence pairs, and the calculated average similarity is normalized to a preset range to obtain the similarity integral of the candidate period. The function expression of the similarity integral is:

[0009] Where, represents the similarity integral of the candidate cycles, Indicates the number of subsequences in the subsequence set, Represents a subsequence The average of the sum of all values ​​in , Represents a subsequence The average of the sum of all values ​​in , represents the dynamic time warping distance, represents the maximum value; L represents the normalization factor.

[0010] Optionally, the obtaining the significance of the candidate period by calculating the similarity integral of the candidate period includes: The mean and standard deviation of the similarity integrals of all candidate periods are calculated and normalized based on the similarity integrals of the candidate periods to obtain the significance of the candidate period. The functional expression of the significance of the candidate period is:

[0011] Where, represents the significance value of the candidate cycle, represents the similarity integral of the candidate cycles, represents the mean of the similarity integrals of all candidate cycles, represents the standard deviation of the similarity integral of all candidate cycles, Represents a minimum constant.

[0012] Optionally, screening out a real cycle from all the candidate cycles according to the significance of the candidate cycles, and determining the real cycle with the maximum significance value as the optimal cycle includes: If the significance value of the candidate period is greater than 0, the candidate period is determined to be a true period; If the significance value of the candidate period is close to 0, the candidate period is determined to be not a real period; From all the candidate cycles satisfying the requirement that the significance value is greater than 0, the true cycle with the largest significance value is selected as the optimal cycle.

[0013] Optionally, the energy storage frequency modulation instruction prediction method further includes calculating the consistency index of the optimal cycle and a method for verifying the optimal cycle, including: Combining the significance of the optimal period and the consistency index of the optimal period with the permutation entropy, the function expression of the consistency index of the optimal period is:

[0014] Where, represents the consistency index of the optimal period, represents the significance value of the optimal period, represents the permutation entropy of the subsequence set corresponding to the optimal period, represents the optimal period; If the consistency index of the optimal period is close to 1, it is determined that the optimal period exists significantly and the pattern is simple; If the consistency index of the optimal period is close to 0, it is determined that the optimal period is not significant or the pattern is too random.

[0015] Optionally, the reconstructing the original frequency modulation instruction to obtain a reconstructed sequence by adopting a chaotic phase injection method based on the consistency index of the optimal period includes: Based on the consistency index of the optimal cycle, the chaotic phase injection method is used to reconstruct the original frequency modulation instruction to obtain a reconstructed sequence. The function expression for reconstructing the original frequency modulation instruction is:

[0016] Where, represents the i-th element in the reconstructed sequence, Represents the i-th element in the original frequency modulation instruction, represents the chaos intensity factor, represents the optimal period, represents the consistency index of the optimal period, represents the sine function, represents pi, represents the periodic perturbation coefficient, represents standard Gaussian noise.

[0017] The present invention also discloses a prediction system, which adopts the above-mentioned energy storage frequency modulation instruction prediction method based on chaotic phase injection, and the prediction system includes: The frequency modulation instruction segmentation module is used to select a group of candidate cycles, and use each of the candidate cycles to segment the collected original frequency modulation instructions to obtain a subsequence set corresponding to the candidate cycles; a sequence alignment module, configured to combine subsequences in the subsequence set in pairs to obtain sequence pairs, calculate the dynamic time warping distance of each sequence pair, and calculate the similarity integral of the candidate period based on the dynamic time warping distance of all the sequence pairs; an optimal period determination module, configured to obtain the significance of the candidate period based on the similarity integral calculation of the candidate period, screen out the real period from all the candidate periods based on the significance of the candidate period, and determine the real period with the maximum significance value as the optimal period; a consistency checking module, configured to calculate and obtain the permutation entropy of the optimal period corresponding to the subsequence set, and calculate a consistency index of the optimal period by combining the significance of the optimal period and the permutation entropy; A sequence reconstruction module, configured to reconstruct the original frequency modulation instruction by adopting a chaotic phase injection method based on the consistency index of the optimal period to obtain a reconstructed sequence; The frequency modulation instruction prediction module is used to input the reconstructed sequence into the pre-trained GRU network and output the predicted value of the original frequency modulation instruction in the preset prediction time length.

[0018] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the above-mentioned energy storage frequency modulation instruction prediction method based on chaotic phase injection is implemented.

[0019] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, characterized in that when the processor executes the computer program, the above-mentioned energy storage frequency modulation instruction prediction method based on chaotic phase injection is implemented.

[0020] Compared with the prior art, the energy storage frequency modulation instruction prediction method and system based on chaotic phase injection provided by the embodiment of the present invention have the following beneficial effects: The dynamic time warping distance is used to accurately quantify the morphological similarity of subsequences. The optimal period consistency index is obtained based on the significance screening of candidate cycles and the calculation of permutation entropy. The chaotic phase injection method is driven to reconstruct the original frequency modulation instructions to generate a low-complexity feature sequence. Combined with the adaptive learning ability of the GRU network, it effectively eliminates timing alignment deviations, decouples chaotic features, and integrates multi-scale characteristics, significantly improving the frequency modulation instruction prediction accuracy, solving the policy oscillation and capacity limit alarm problems caused by traditional methods, and enhancing the regulation reliability and economy of the energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments, in which: Figure 1 A schematic diagram of the steps of the energy storage frequency modulation instruction prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. Now, in conjunction with the accompanying drawings, the preferred embodiments of the present invention will be described in detail.

[0023] The present invention discloses a method for predicting energy storage frequency modulation instructions based on chaotic phase injection, such as Figure 1 Shown, including: S1. Select a group of candidate cycles, and use each candidate cycle to segment the collected original frequency modulation instructions to obtain a set of subsequences corresponding to the candidate cycles. S2. Combine the subsequences in the subsequence set into pairs to obtain sequence pairs, calculate the dynamic time warping distance of each sequence pair, and calculate the similarity integral of the candidate period based on the dynamic time warping distance of all sequence pairs; S3. Obtain the significance of the candidate periods based on the similarity integral calculation of the candidate periods, select the real periods from all candidate periods based on the significance of the candidate periods, and determine the real period with the maximum significance value as the optimal period; S4. Calculate the permutation entropy of the subsequence set corresponding to the optimal period, and calculate the consistency index of the optimal period by combining the significance of the optimal period and the permutation entropy; S5. Based on the consistency index of the optimal period, the chaotic phase injection method is used to reconstruct the original frequency modulation instruction to obtain a reconstructed sequence; S6. Input the reconstructed sequence into the pre-trained GRU network, and output the predicted value of the original frequency modulation instruction in the preset prediction time length.

[0024] Through the implementation of the above-mentioned energy storage frequency modulation instruction prediction method embodiment, the dynamic time warping distance (DTW) is used to accurately quantify the nonlinear morphological similarity of all sequence pairs in the subsequence set formed by candidate cycle segmentation, overcoming the insensitivity of traditional cycle detection to time axis expansion / compression, and significantly improving the cycle capture capability of non-stationary frequency modulation instructions. The statistical significance of the candidate cycle is calculated based on the similarity integral, and the risk of misjudgment caused by noise interference is eliminated through standardization. The true cycle is screened and the optimal cycle with the maximum significance value is locked. The cycle consistency index is calculated by combining the permutation entropy of the subsequence set corresponding to the optimal cycle. The entropy weighting mechanism is used to ensure that the cycle pattern satisfies both statistical significance and low complexity characteristics, completely eliminating the interference of complex random patterns on cycle selection. A chaotic phase injection method is then used to adaptively reconstruct the original frequency modulation instructions, generating a reconstructed sequence that combines the characteristics of the main cycle with those of chaotic fluctuations. Finally, a pre-trained GRU network (Gated Recurrent Unit) is used to achieve multi-timescale coordinated prediction. This effectively addresses the policy oscillation caused by timing misalignment, prediction distortion in sudden bursts caused by insufficient decoupling of chaotic features, and capacity over-limit alarms caused by a lack of multi-timescale coordination in traditional methods. The reconstructed sequence significantly reduces the nonlinear complexity of the original frequency modulation instructions, enabling the GRU network to accurately decouple the chaotic characteristics of the coupling between grid inertia and renewable energy penetration. This simultaneously improves the coordination between second-level real-time response and minute-level State of Charge (SOC) planning, eliminates the dynamic lag between frequency modulation instructions and energy storage response, and ensures that the energy storage system accurately tracks grid frequency regulation requirements in scenarios with a high renewable energy penetration. This avoids battery life degradation caused by frequent switching of charge and discharge strategies, improves regulation capacity matching accuracy and response qualification rate, and thus maintains grid frequency stability while extending the service life of energy storage equipment and increasing frequency regulation compensation benefits.

[0025] Furthermore, the collected original frequency modulation instructions are segmented using each candidate cycle to obtain a set of subsequences corresponding to the candidate cycles, including: The number of divisions of the original FM instruction is determined according to the total number of sampling times of the original FM instruction and the candidate cycle. The function expression of the number of divisions of the original FM instruction is:

[0026] Where K represents the number of subsequences after segmentation, N represents the total number of sampling times of the original frequency modulation instruction, T represents the candidate cycle, Indicates rounding down; According to the determined number of divisions, the original frequency modulation instruction is divided into a corresponding number of subsequences using candidate cycles, and the divided subsequences are arranged in a division order to form a subsequence set.

[0027] By implementing the above-mentioned energy storage frequency modulation instruction prediction method embodiment, the number of subsequences is accurately and dynamically determined according to the relevant function in the segmentation stage, wherein the rounding down operation completely avoids the problem of incomplete subsequences at the tail caused by non-integer cycles, ensuring that the subsequence set structure generated by each candidate cycle T is complete, such as . This quantized segmentation mechanism ensures that each subsequence strictly contains T continuous sampling points (except for the subsequence with less than T at the end, which automatically adapts to N length), eliminating the fixed length deviation of the traditional sliding window from the source. The set of subsequences arranged in the order of segmentation completely retains the phase continuity of the frequency modulation instruction in the time dimension, and constructs an alignment benchmark for the subsequent dynamic time warping distance calculation. Compared with the edge point truncation distortion caused by traditional equal-length segmentation, the method of this embodiment ensures that each subsequence independently carries the local timing pattern through dynamic segmentation and rounding down control of the candidate period T, so that subsequent period detection can identify the short-cycle frequency modulation characteristics caused by inertia differences in the new energy high-penetration power grid, and ultimately significantly reduce the action delay and strategy switching frequency of the energy storage response.

[0028] For example, given the original frequency modulation instruction P=[X1,X2,X3……X N ], N is the number of sampling times; select a set of candidate cycles It is between 7 and 15. For each candidate cycle T, the original frequency modulation instruction is divided into K subsequences to obtain a subsequence set .

[0029] Furthermore, the dynamic time warping distance of each sequence pair is calculated, and the similarity integral of the candidate period is calculated based on the dynamic time warping distance of all sequence pairs, including: For each sequence, the dynamic time warping distance is obtained by minimizing the cumulative path distance through dynamic programming; The average similarity is calculated based on the dynamic time warping distance of all sequence pairs, and the calculated average similarity is normalized to the similarity integral of the candidate period within the preset range. The function expression of the similarity integral is:

[0030] Where, represents the similarity integral of the candidate cycles, Indicates the number of subsequences in the subsequence set, Represents a subsequence The average of the sum of all values ​​in , Represents a subsequence The average of the sum of all values ​​in , represents the dynamic time warping distance, represents the maximum value; L represents the normalization factor.

[0031] By implementing the above energy storage frequency modulation instruction prediction method embodiment, dynamic time warping distance DTW is used to replace the traditional linear correlation calculation to solve the period detection problem under nonlinear time deformation. That is, for each candidate period T, the sequence pairs in the subsequence set generated by segmentation are , strictly based on the dynamic programming principle to minimize the cumulative path distance to obtain the DTW value, effectively overcoming the inadaptability of fixed phase matching to time axis expansion / compression. On this basis, the similarity integral function expression is established, and the dynamic time warping distance of 2 / K(K-1) non-repeating sequence pairs is normalized through a double summation mechanism. The dynamic time warping distance on the numerator side of the core operator accurately quantifies the sequence morphological similarity, and the denominator side uses the normalization factor L and the average value of the sum of all subsequence values ​​(i.e. , ) constitutes an adaptive scaling benchmark, which not only eliminates interference from amplitude differences but also enhances compatibility with subsequences of unequal lengths through the maximum value function. This enables the computational process to simultaneously perform multi-scale periodic analysis. For example, the above operations are performed in parallel for 7-15 candidate periods T, and each T independently generates K subsequences. By traversing the similarity integral C(T) under different T values, the system covers potential periodic patterns from seconds to minutes. Thus, a dual synergy is formed through dynamic time warping distance calculation and multi-candidate period traversal. The former ensures the accuracy of morphological similarity measurement at each scale, and the latter avoids feature omissions caused by single-period presets. Together, the similarity integral C(T) of the candidate period becomes a robust indicator to characterize period significance, thereby solving the problem of false period detection in nonlinear time deformation scenarios caused by traditional methods.

[0032] Furthermore, the significance of the candidate period is obtained based on the similarity integral calculation of the candidate period, including: The mean and standard deviation of the similarity integrals of all candidate periods are calculated and normalized based on the similarity integrals of the candidate periods to obtain the significance of the candidate periods. The functional expression of the significance of the candidate periods is:

[0033] Where, represents the significance value of the candidate cycle, represents the similarity integral of the candidate cycles, represents the mean of the similarity integrals of all candidate cycles, represents the standard deviation of the similarity integral of all candidate cycles, Represents a minimum constant.

[0034] Furthermore, the true cycle is screened out from all candidate cycles according to the significance of the candidate cycles, and the true cycle with the maximum significance value is determined as the optimal cycle, including: If the significance value of the candidate period is greater than 0, the candidate period is judged to be the true period; If the significance value of a candidate cycle is close to 0, the candidate cycle is judged to be not a real cycle; From all candidate cycles that satisfy the significance value greater than 0, the true cycle with the largest significance value is selected as the optimal cycle.

[0035] By implementing the above energy storage frequency modulation instruction prediction method embodiment, the average value of the similarity integral C(T) of all candidate cycles is obtained. is the random noise baseline level, standard deviation Quantify the overall fluctuation intensity and use standardization to convert the single-period C(T) into a significance value PSS(T). This breaks through the limitations of traditional absolute thresholds, i.e., the denominator + Constitute an adaptive noise filtering layer to make molecules - Only periodic signals that significantly exceed the population mean are retained. ) is automatically determined as invalid noise, eliminating the empirical dependence of manually setting fixed thresholds. A statistical significance criterion is established based on the screening condition of PSS(T)>0: only when PSS(T) is much greater than 0, the similarity of the cycle is determined to be significantly higher than random fluctuations and may be a real cycle; when PSS(T) is close to 0, the periodicity is determined to be equivalent to the noise level and is not an approximate cycle of the sequence. This eliminates the possibility that the cycle is lower than The random fluctuation interference is eliminated, and on this basis, the optimal cycle is locked with the maximum value of PSS(T), forming a double statistical verification closed loop. This method significantly improves the robustness to short-term fluctuation noise in the frequency modulation instruction: for example, when a candidate cycle has an abnormal increase in similarity integral due to a sudden load change, Synchronous expansion keeps PSS(T) stable; conversely, in low noise scenarios Shrinkage enhances the cycle resolution sensitivity to avoid the dual defects of the fixed threshold scheme, which is missing valid cycles in high-noise environments and mistaking false cycles in low-noise environments.

[0036] For example, for each candidate cycle Calculate the corresponding similarity score , calculate all candidate cycles The mean and standard deviation of each candidate cycle are normalized to obtain the significance of each candidate cycle. , select the true period with the largest significance value as the optimal period, that is The corresponding candidate period is the optimal period.

[0037] Furthermore, the energy storage frequency modulation instruction prediction method also includes a method for calculating the consistency index of the optimal cycle and verifying the optimal cycle, including: Combining the significance of the optimal period and the consistency index of the permutation entropy to the optimal period, the function expression of the optimal period consistency index is:

[0038] Where, represents the consistency index of the optimal period, represents the significance value of the optimal period, represents the permutation entropy of the subsequence set corresponding to the optimal period, represents the optimal period; If the consistency index of the optimal cycle is close to 1, it is determined that the optimal cycle exists significantly and the pattern is simple; If the consistency index of the optimal period is close to 0, it is judged that the optimal period is not significant or the pattern is too random.

[0039] By implementing the above energy storage frequency modulation instruction prediction method embodiment, the significance value of the optimal cycle is As the base value, through the reciprocal operator 1 / (1+ ( )) The optimal period corresponds to the permutation entropy of the subsequence set ( ) is converted into an inverse weight coefficient, forming a dual-constraint fusion logic: when the permutation entropy approaches 0 (the subsequence pattern is highly ordered), the weight coefficient approaches 1, making PCI ≈ To preserve the significance of the original cycle; when the permutation entropy increases significantly (subsequence complexity increases), the weight coefficient decays geometrically, forcing PCI to converge to 0. This method ensures that the cycle consistency index satisfies The statistical significance of the representation and ( ) The low complexity of the measurement accurately implements the defined judgment criteria: if PCI→1, then Close to the theoretical peak and ( )≈0 (the optimal period is significant and the pattern is simple); if PCI→0, it reflects Suppressed by high entropy values ​​(optimal period is not significant or the pattern is too random). Based on the entropy weighting mechanism, we ensure that the periodic pattern satisfies both significance and low complexity to quantify the overall periodicity.

[0040] Furthermore, based on the consistency index of the optimal period, the chaotic phase injection method is used to reconstruct the original frequency modulation instruction to obtain a reconstructed sequence, including: Based on the consistency index of the optimal cycle, the chaotic phase injection method is used to reconstruct the original frequency modulation instruction to obtain the reconstructed sequence. The function expression for reconstructing the original frequency modulation instruction is:

[0041] Where, represents the i-th element in the reconstructed sequence, Represents the i-th element in the original frequency modulation instruction, represents the chaos intensity factor, represents the optimal period, represents the consistency index of the optimal period, represents the sine function, represents pi, represents the periodic perturbation coefficient, represents standard Gaussian noise.

[0042] By implementing the above energy storage frequency modulation instruction prediction method embodiment, the sine function sin() is strictly based on the optimal period. Generate a periodic modulation signal for the reference frequency, and the pi π is coupled with the index i to form a phase conversion operator, which accurately retains the main period characteristics of the power grid frequency regulation instruction. And based on three synergistic mechanisms: the optimal period consistency index PCI is directly related to the chaos intensity factor A product constraint is formed to make the chaotic intensity dynamically decay with the PCI value, that is, when PCI→1 (strong significant cycle), the chaotic interference is suppressed, and when PCI→0 (weak cycle), the phase disturbance is enhanced, forming an adaptive anti-overfitting barrier; standard Gaussian noise By periodic perturbation coefficient Linear coupling to the denominator produces ±30% cycle length dynamic jitter to overcome the sensitivity of the fixed cycle preset to grid inertia fluctuations; the product term The nonlinear superposition relationship between the original frequency modulation instruction and the chaotic phase is established, which not only maintains the amplitude distribution characteristics but also injects spectrum disturbance. Preserve static characteristics, decouple dynamic chaotic characteristics of sinusoidal modulation signals, and standard Gaussian noise terms Generate frequency domain anti-aliasing noise The three work together to generate a low-complexity reconstruction sequence , significantly improving the GRU network's predictive adaptability to non-steady-state frequency modulation instructions, and completely eliminating the traditional method's strategy oscillation caused by timing alignment deviation and capacity limit alarms caused by the lack of multi-scale coordination.

[0043] For example, the original frequency modulation instruction P=[X1,X2,X3……X N ]Refactored to =[X1 、 ,X2 、 ,X3、 ...X N 、 ],Bundle =[X1 、 ,X2 、 ,X3 、 ...X N 、 ] is put into the pre-trained GRU network for prediction and the predicted value [X N+1 ,X N +2,X N+3 ...X N+S ], the predicted value is the predicted value of the original frequency modulation instruction.

[0044] In order to further verify the advantages of the present invention, the method of the embodiment of the present invention and the GRU network direct prediction method are used to predict the frequency modulation sequence, and the four evaluation indicators of MAE (Mean Absolute Error), SSE (Sum of Squares Error), RMSE (Root Mean Square Error), and MAPE (Mean Absolute Percentage Error) are used to evaluate the above prediction results. The selection of evaluation indicators can be referred to the evaluation indicator table as shown below: Table 1 Evaluation index table

[0045] As shown in Table 1, among the four evaluation indicators, N represents the sample size. and Represent the actual value and predicted value at time n respectively. The results of evaluating the above prediction results using four evaluation indicators are shown in the FM sequence prediction table: Table 2 FM sequence prediction table

[0046] As shown in Table 2, FM sequence 1 is derived from the FM data of a power plant in Hulunbuir from 0:00 to 20:00 on December 1, 2024; FM sequence 2 is derived from the FM data of a power plant in Hulunbuir from 0:00 to 22:00 on December 2, 2024. The evaluation results in Table 2 show that all four evaluation indicators have been reduced, indicating that the decomposition method proposed in this invention can reduce the nonlinearity of the original sequence, further improve the prediction accuracy, and effectively help power plants improve their frequency modulation response capabilities, further increasing their profits.

[0047] The present invention also discloses a prediction system, which adopts the above-mentioned energy storage frequency modulation instruction prediction method based on chaotic phase injection, and the prediction system includes: The FM instruction segmentation module is used to select a group of candidate cycles and segment the collected original FM instructions using each candidate cycle to obtain a set of subsequences corresponding to the candidate cycles. The sequence alignment module is used to combine the subsequences in the subsequence set into sequence pairs, calculate the dynamic time warping distance of each sequence pair, and calculate the similarity integral of the candidate period based on the dynamic time warping distance of all sequence pairs; The optimal cycle determination module is used to obtain the significance of the candidate cycles based on the similarity integral calculation of the candidate cycles, screen out the real cycles from all candidate cycles based on the significance of the candidate cycles, and determine the real cycle with the maximum significance value as the optimal cycle; The consistency check module is used to calculate the permutation entropy of the subsequence set corresponding to the optimal period, and to calculate the consistency index of the optimal period by combining the significance of the optimal period and the permutation entropy; The sequence reconstruction module is used to reconstruct the original frequency modulation instruction based on the consistency index of the optimal period and obtain the reconstructed sequence by using the chaotic phase injection method; The FM instruction prediction module is used to input the reconstructed sequence into the pre-trained GRU network and output the predicted value of the original FM instruction at the preset prediction duration.

[0048] The present invention also discloses a computer-readable storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, the above-mentioned energy storage frequency modulation instruction prediction method based on chaotic phase injection is implemented.

[0049] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, characterized in that the processor implements the above-mentioned energy storage frequency modulation instruction prediction method based on chaotic phase injection when executing the computer program.

[0050] The present invention is described based on flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to specific embodiments. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0051] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0052] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0053] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Those skilled in the art may modify the technical solutions described in the above embodiments, or replace some of the technical features therein with equivalents; and all these modifications and replacements should fall within the scope of protection of the present invention.

Claims

1. A method for predicting energy storage frequency modulation instructions based on chaotic phase injection, characterized in that: The energy storage frequency modulation instruction prediction method includes: Selecting a group of candidate cycles, and using each of the candidate cycles to segment the collected original frequency modulation instructions, to obtain a set of subsequences corresponding to the candidate cycles; Combining subsequences in the subsequence set in pairs to obtain sequence pairs, calculating the dynamic time warping distance of each of the sequence pairs, and calculating the similarity integral of the candidate period based on the dynamic time warping distances of all the sequence pairs; Obtaining significance of the candidate periods according to the similarity integral calculation of the candidate periods, screening out real periods from all the candidate periods according to the significance of the candidate periods, and determining the real period with the maximum significance value as the optimal period; Calculating and obtaining the permutation entropy of the subsequence set corresponding to the optimal period, and calculating and obtaining the consistency index of the optimal period by combining the significance of the optimal period and the permutation entropy; Based on the consistency index of the optimal period, a chaotic phase injection method is used to reconstruct the original frequency modulation instruction to obtain a reconstructed sequence; The reconstructed sequence is input into a pre-trained GRU network, and the output is a predicted value of the original frequency modulation instruction at a preset prediction duration.

2. The energy storage frequency modulation instruction prediction method based on chaotic phase injection according to claim 1 is characterized in that: The collected original frequency modulation instructions are divided respectively by using each candidate cycle to obtain a subsequence set corresponding to the candidate cycle, including: The number of divisions of the original frequency modulation instruction is determined according to the total number of sampling times of the original frequency modulation instruction and the candidate cycle. The function expression of the number of divisions of the original frequency modulation instruction is: Where K represents the number of subsequences after segmentation, N represents the total number of sampling times of the original frequency modulation instruction, T represents the candidate cycle, Indicates rounding down; According to the determined number of divisions, the original frequency modulation instruction is divided into a corresponding number of subsequences using the candidate cycles, and the divided subsequences are arranged in a division order to form the subsequence set.

3. The energy storage frequency modulation instruction prediction method based on chaotic phase injection according to claim 1 is characterized in that: Calculating the dynamic time warping distance of each of the sequence pairs and obtaining the similarity integral of the candidate period based on the dynamic time warping distances of all the sequence pairs includes: For each of the sequences, a dynamic time warping distance is obtained by minimizing the cumulative path distance through dynamic programming; The average similarity is calculated based on the dynamic time warping distance of all the sequence pairs, and the calculated average similarity is normalized to a preset range to obtain the similarity integral of the candidate period. The function expression of the similarity integral is: Where, represents the similarity integral of the candidate cycles, Indicates the number of subsequences in the subsequence set, Represents a subsequence The average of the sum of all values ​​in , Represents a subsequence The average of the sum of all values ​​in , represents the dynamic time warping distance, represents the maximum value; L represents the normalization factor.

4. The energy storage frequency modulation instruction prediction method based on chaotic phase injection according to claim 1 is characterized in that: The obtaining of the significance of the candidate period by calculating the similarity integral of the candidate period includes: The mean and standard deviation of the similarity integrals of all candidate periods are calculated and normalized based on the similarity integrals of the candidate periods to obtain the significance of the candidate period. The functional expression of the significance of the candidate period is: Where, represents the significance value of the candidate cycle, represents the similarity integral of the candidate cycles, represents the mean of the similarity integrals of all candidate cycles, represents the standard deviation of the similarity integral of all candidate cycles, Represents a minimum constant.

5. The energy storage frequency modulation instruction prediction method based on chaotic phase injection according to claim 4 is characterized in that: The step of selecting a real cycle from all the candidate cycles according to the significance of the candidate cycles and determining the real cycle with the maximum significance value as the optimal cycle includes: If the significance value of the candidate period is greater than 0, the candidate period is determined to be a true period; If the significance value of the candidate period is close to 0, the candidate period is determined to be not a real period; From all the candidate cycles satisfying the requirement that the significance value is greater than 0, the true cycle with the largest significance value is selected as the optimal cycle.

6. The energy storage frequency modulation instruction prediction method based on chaotic phase injection according to claim 1 is characterized in that: The energy storage frequency modulation instruction prediction method further includes calculating the consistency index of the optimal cycle and a method for verifying the optimal cycle, including: Combining the significance of the optimal period and the consistency index of the optimal period with the permutation entropy, the function expression of the consistency index of the optimal period is: Where, represents the consistency index of the optimal period, represents the significance value of the optimal period, represents the permutation entropy of the subsequence set corresponding to the optimal period, represents the optimal period; If the consistency index of the optimal period is close to 1, it is determined that the optimal period exists significantly and the pattern is simple; If the consistency index of the optimal period is close to 0, it is determined that the optimal period is not significant or the pattern is too random.

7. The energy storage frequency modulation instruction prediction method based on chaotic phase injection according to claim 1 is characterized in that: The method of reconstructing the original frequency modulation instruction based on the consistency index of the optimal period by adopting a chaotic phase injection method to obtain a reconstructed sequence includes: Based on the consistency index of the optimal cycle, the chaotic phase injection method is used to reconstruct the original frequency modulation instruction to obtain a reconstructed sequence. The function expression for reconstructing the original frequency modulation instruction is: Where, represents the i-th element in the reconstructed sequence, Represents the i-th element in the original frequency modulation instruction, represents the chaos intensity factor, represents the optimal period, represents the consistency index of the optimal period, represents the sine function, represents pi, represents the periodic perturbation coefficient, represents standard Gaussian noise.

8. A prediction system, using the energy storage frequency modulation instruction prediction method based on chaotic phase injection according to any one of claims 1 to 7, characterized in that: The prediction system includes: The frequency modulation instruction segmentation module is used to select a group of candidate cycles, and use each of the candidate cycles to segment the collected original frequency modulation instructions to obtain a subsequence set corresponding to the candidate cycles; a sequence alignment module, configured to combine subsequences in the subsequence set in pairs to obtain sequence pairs, calculate the dynamic time warping distance of each sequence pair, and calculate the similarity integral of the candidate period based on the dynamic time warping distance of all the sequence pairs; an optimal period determination module, configured to obtain the significance of the candidate period based on the similarity integral calculation of the candidate period, screen out the real period from all the candidate periods based on the significance of the candidate period, and determine the real period with the maximum significance value as the optimal period; a consistency checking module, configured to calculate and obtain the permutation entropy of the optimal period corresponding to the subsequence set, and calculate a consistency index of the optimal period by combining the significance of the optimal period and the permutation entropy; A sequence reconstruction module, configured to reconstruct the original frequency modulation instruction by adopting a chaotic phase injection method based on the consistency index of the optimal period to obtain a reconstructed sequence; The frequency modulation instruction prediction module is used to input the reconstructed sequence into the pre-trained GRU network and output the predicted value of the original frequency modulation instruction in the preset prediction time length.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the energy storage frequency modulation instruction prediction method based on chaotic phase injection according to any one of claims 1 to 7 is implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the energy storage frequency modulation instruction prediction method based on chaotic phase injection according to any one of claims 1 to 7 is implemented.

Images