Method for automated scheduling based on load and distributed energy fluctuations

Abstract

The present application relates to a method for automatic scheduling based on load and distributed energy fluctuation, comprising the following steps: data acquisition: acquiring load prediction time series data and distributed energy generation prediction time series data, generating an initial scheduling plan based on the load prediction time series data and the distributed energy generation prediction time series data, the initial scheduling plan including a conventional unit output plan and a storage energy charging and discharging plan; data acquisition: real-time acquisition of load actual power time series data and distributed energy actual power time series data; fluctuation decomposition: adaptive variational mode decomposition is performed on the load actual power time series data and the distributed energy actual power time series data respectively, and load fluctuation components and distributed energy fluctuation components are extracted; index calculation: according to the load fluctuation components and the distributed energy fluctuation components; the present application can accurately quantify fluctuation dynamic scheduling to improve the stability and economy of the power grid.

Description

Technical Field

[0001] This invention relates to the field of scheduling methods, and more specifically to a method for automated scheduling based on load and distributed energy fluctuations. Background Technology

[0002] With the gradual integration of distributed energy resources into the power grid, both load power and distributed energy generation power exhibit significant time-series fluctuations. The coupling effect of these dual fluctuations presents numerous technical challenges for automated grid dispatching, becoming a major pain point in existing dispatching technologies. Traditional dispatching methods often employ fixed-parameter decomposition to extract fluctuation components from power time-series data, which easily leads to over- or under-decomposition of signals. This fails to accurately capture the true fluctuation characteristics of load and distributed energy, creating potential data biases for subsequent fluctuation analysis and dispatching decisions. Furthermore, the quantification of overall system fluctuations only considers the independent fluctuation intensity of load or distributed energy, neglecting to introduce cross-coupling terms characterizing the synchronicity of their fluctuations. This fails to reflect the superimposed amplification effect resulting from the interaction of load and distributed energy in amplitude and phase, leading to distorted comprehensive fluctuation quantification results and hindering the provision of scientifically accurate quantitative basis for dispatching triggering and adjustment. According to reports, the trigger thresholds for dynamic dispatch corrections are mostly fixed values, which cannot be adaptively adjusted based on real-time operating conditions such as grid frequency deviation and energy storage state of charge. This easily leads to problems such as thresholds that are too high, resulting in untimely fluctuation smoothing, and thresholds that are too low, causing frequent and ineffective adjustments by conventional units and energy storage, resulting in extremely poor adaptability. Furthermore, the decision-making process for dispatch plan corrections often suffers from a single input dimension, failing to comprehensively consider the overall system fluctuation level, the actual adjustment capability of energy storage, and the fluctuation correlation characteristics of load and distributed energy resources. Moreover, it often employs a single-objective optimization strategy, making it difficult to simultaneously minimize grid operating costs and maximize power fluctuation smoothing effects. Even when using fuzzy decision-making mechanisms, the rules... The library also lacks precise algorithmic optimization, making it difficult for the output to approximate the optimal solution under actual operating conditions. Traditional model predictive control frameworks often use fixed-weight objective functions in scheduling corrections, failing to dynamically adjust optimization priorities based on the severity of system fluctuations. When fluctuations are severe, they struggle to enhance grid-connected power mitigation; when fluctuations are mild, they overemphasize stability while neglecting economic efficiency. Furthermore, the framework lacks proactive prediction and feedback correction mechanisms for future fluctuations, relying solely on historical and current operating data for scheduling predictions. This insufficient awareness of future fluctuation uncertainties leads to significant deviations between the predicted output curve and actual operating conditions, greatly reducing the accuracy of rolling optimization. Overall... It is argued that existing power grid dispatching technologies have failed to form a complete closed loop from accurate extraction of fluctuation components, scientific quantification of comprehensive fluctuations, dynamic adaptation of trigger thresholds to intelligent correction of dispatching plans. The response to load and distributed energy fluctuations lacks real-time, targeted, and forward-looking capabilities. The coordinated dispatching effect of conventional units and energy storage systems is poor, and over- or under-regulation is prone to occur. It is neither able to effectively smooth out system power fluctuations to ensure stable grid operation, nor can it easily increase operating costs and mechanical losses due to frequent and large-scale equipment adjustments. It is difficult to adapt to the dispatching needs of power grids with a high proportion of distributed energy. Therefore, a method for automated dispatching based on load and distributed energy fluctuations is proposed. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method for automated scheduling based on load and distributed energy fluctuations. Data acquisition includes acquiring load forecast time-series data and distributed energy generation forecast time-series data, and generating an initial scheduling plan based on the load forecast time-series data and distributed energy generation forecast time-series data. The initial scheduling plan includes a conventional unit output plan and an energy storage charging and discharging plan.

[0004] Data acquisition: Real-time acquisition of actual load power time-series data and distributed energy power time-series data;

[0005] Fluctuation decomposition: Adaptive variational mode decomposition is performed on the actual power time series data of load and the actual power time series data of distributed energy to extract load fluctuation components and distributed energy fluctuation components.

[0006] Index Calculation: Based on the load fluctuation component and the distributed energy fluctuation component, a comprehensive fluctuation index is obtained through a coupled fluctuation index calculation model. The coupled fluctuation index calculation model includes cross-coupling terms used to characterize the synchronicity of load and distributed energy fluctuations.

[0007] Threshold judgment: Determine whether the comprehensive volatility index exceeds the trigger threshold dynamically generated based on the real-time operating status. If it does, a dynamic correction process is triggered.

[0008] Corrective scheduling: During the dynamic correction process, based on the comprehensive fluctuation index and the current energy storage charge state, a multi-objective fuzzy decision-making mechanism is adopted to adjust the initial scheduling plan, generate corrective scheduling instructions, and issue them for execution.

[0009] Furthermore, the specific process of adaptive variational mode decomposition for the actual power time series data of loads and the actual power time series data of distributed energy sources includes:

[0010] For the actual power time series data f of the load load (t) and time-series data of actual power of distributed energy sources f der (t) Construct variational constraint models respectively, and use the whale optimization algorithm to optimize the penalty factor α and the number of modal components K of variational mode decomposition;

[0011] The optimization process takes minimizing the envelope entropy as the objective function, calculates the envelope entropy value of each individual, and obtains the optimal parameter combination [αopt, Kopt] through iterative updates;

[0012] The original signal is decomposed based on optimal parameters to obtain a series of intrinsic mode functions (EMFs). The EMFs with center frequencies higher than the preset cutoff frequency are then reconstructed into load fluctuation components W. load (t) and distributed energy fluctuation component W der (t).

[0013] Furthermore, the specific process of obtaining the comprehensive volatility index through the coupled volatility index calculation model includes:

[0014] Within a preset sliding time window, Hilbert transforms are performed on the load fluctuation component and the distributed energy fluctuation component respectively to obtain the instantaneous amplitude function and instantaneous phase function at each moment within the time window;

[0015] The instantaneous amplitude ratio of the load fluctuation component and the distributed energy fluctuation component at each moment within the time window is calculated, and the instantaneous amplitude ratio at all moments within the time window is integrated to quantify the relative change in amplitude between the two, thus obtaining the first coupling coefficient C1.

[0016] The instantaneous phase difference between the load fluctuation component and the distributed energy fluctuation component at each moment within the time window is calculated, and the variance of the instantaneous phase difference at all moments within the time window is statistically analyzed to measure the synchronicity or misalignment of the two in the fluctuation time sequence, and the second coupling coefficient C2 is obtained.

[0017] The relative load fluctuation intensity is obtained by calculating the standard deviation of the load fluctuation component within the time window and dividing it by the base power value. Similarly, the relative distributed energy fluctuation intensity is obtained by calculating the standard deviation of the distributed energy fluctuation component within the time window and dividing it by the base power value. The square root of the sum of the squares of both components is then multiplied by an amplification factor consisting of a linear combination of the first and second coupling coefficients to obtain the comprehensive fluctuation index. The calculation formula is as follows:

[0018] ;

[0019] in, and These are the standard deviations of load and distributed energy fluctuation components, respectively. and γ1 and γ2 are preset coupling strength coefficients, and the introduction of cross-coupling terms characterizes the superposition amplification effect of the interaction between load and distributed energy fluctuations in amplitude and phase on the overall system volatility.

[0020] Furthermore, the trigger threshold is a dynamically generated threshold based on the real-time running status, and the specific acquisition process includes:

[0021] Obtain the current power grid frequency deviation And the state of charge (SOC) of energy storage, with the current grid frequency deviation Using the energy storage state of charge (SOC) as input variables, the frequency deviation adjustment factor k is directly output through a pre-generated two-dimensional fuzzy controller. f and the state of charge deviation adjustment factor k s ;

[0022] Based on the basic threshold T base Frequency deviation adjustment factor k f State of charge deviation adjustment factor k s Current power grid frequency deviation and the energy storage state of charge (SOC) and reference SOC ref Deviation | SOC-SOC ref Calculate the trigger threshold Specifically:

[0023] ;

[0024] Among them, SOC ref As a state of charge reference value, the threshold is adaptively changed according to the grid frequency health status and energy storage regulation capability.

[0025] Furthermore, the adjustment of the initial scheduling plan using a multi-objective fuzzy decision-making mechanism includes:

[0026] The comprehensive volatility index I cwi The state of charge (SOC) of energy storage and the correlation coefficient ρ between load fluctuation components and distributed energy fluctuation components are used as three-dimensional input variables into the multi-input fuzzy inference system.

[0027] Establish a fuzzy rule base with the dual objectives of minimizing system operating costs and maximizing power fluctuation mitigation, and output the power adjustment amount of conventional generating units. and energy storage power adjustment amount ;

[0028] The establishment of the fuzzy rule base is based on offline optimization. The parameters of the membership function are encoded and optimized through a genetic algorithm, so that the output of the fuzzy inference system approaches the Pareto optimal frontier in typical historical scenarios.

[0029] Furthermore, the correlation coefficient ρ is obtained by calculating the Pearson correlation coefficient between the load fluctuation component and the distributed energy fluctuation component. Specifically, the deviation of the load fluctuation component from its mean at each sampling time is calculated, multiplied by the deviation of the distributed energy fluctuation component from its mean, and the sum of these products is divided by the product of the standard deviations of the load fluctuation component and the distributed energy fluctuation component to obtain the correlation coefficient, thereby quantifying the degree of linear correlation between the two fluctuations. The calculation formula is as follows:

[0030] ;

[0031] Let be the load fluctuation component at time t. Let be the distributed energy fluctuation component at time t, and both are obtained by adaptive variational mode decomposition;

[0032] This represents the average value of the load fluctuation component over the time window. is the mean of the distributed energy fluctuation component within the time window; n is the total number of sampling points within the time window;

[0033] The correlation coefficient is used to determine whether load and distributed energy fluctuations cancel each other out or overlap during fuzzy inference. When ρ is negative, the mutual cancellation between fluctuations is used to reduce energy storage action. When ρ is positive, the energy storage compensation is increased.

[0034] Furthermore, the dynamic correction process employs an improved model prediction control framework, constructing a time-domain variable-weight objective function based on a comprehensive volatility index. The specific process includes:

[0035] Within the prediction time domain, the comprehensive volatility index is used as an adjustment factor for the weighting coefficients to construct a rolling optimization objective function, which consists of three parts:

[0036] The first part is the grid-connected power P. grid (k) and power reference value P ref The square of the difference, multiplied by a dynamic weight, is used to suppress grid-connected power fluctuations; the second part is the power adjustment amount for conventional units. The square of the product of the fixed weights The first part is used to limit the unit's regulation range; the third part consists of the energy storage state of charge (SOC(k)) and the state of charge reference value (SOC). ref The square of the difference multiplied by a fixed weight This is used to maintain the healthy state of energy storage, where SOC(k) is the SOC value of the energy storage at time k; the dynamic weight changes linearly with the comprehensive fluctuation index, and the calculation formula is as follows:

[0037] ;

[0038] Among them, the weighting coefficient , With the base weight, β as the adjustment coefficient, and Np as the prediction time domain length, this system automatically enhances the weight for suppressing grid-connected power fluctuations when fluctuations are severe, while focusing on economic operation when fluctuations are mild.

[0039] Furthermore, the model predictive control framework also includes a feedback correction stage based on fluctuation component prediction, the specific acquisition process of which includes:

[0040] The load fluctuation component W obtained from real-time decomposition load (t) and distributed energy fluctuation component W der (t) is input into a long short-term memory neural network to predict the trend of the fluctuation components in the future time domain;

[0041] The predicted fluctuation component trend is superimposed on the predicted output curve generated based on the initial scheduling plan, and used as a correction term for the prediction model in the model predictive control framework to achieve advance perception and pre-adjustment of future uncertainties.

[0042] The present invention has the following advantages over the prior art:

[0043] By employing adaptive variational mode decomposition combined with the whale optimization algorithm on actual power time-series data, the optimal parameters are obtained with the goal of minimizing envelope entropy. This accurately extracts the fluctuation components of load and distributed energy, laying a precise data foundation for subsequent fluctuation analysis. The coupled fluctuation index calculation model introduces a cross-coupling term characterizing the synchronicity of load and distributed energy fluctuations. Combined with the instantaneous amplitude and phase characteristics obtained from Hilbert transform, the coupling coefficient is calculated. By comprehensively considering the interaction of the relative fluctuation intensity, amplitude, and phase of the two, the overall fluctuation effect of the system can be accurately quantified, accurately reflecting the superposition and amplification of fluctuation effects. The trigger threshold is based on grid frequency deviation and energy storage state of charge. By dynamically generating adjustment factors through a two-dimensional fuzzy controller, the threshold can adaptively change with the grid frequency health status and energy storage regulation capacity, avoiding the limitations of fixed thresholds and making the triggering of the dynamic correction process more closely match the real-time operating status of the grid. The multi-objective fuzzy decision-making mechanism uses the comprehensive fluctuation index, energy storage state of charge, and Pearson correlation coefficient as three-dimensional inputs. The fuzzy rule base is optimized by a genetic algorithm to approach the Pareto optimal frontier under historical typical scenarios. It can balance minimizing system operating costs and maximizing power fluctuation smoothing effects. It can also judge the mutual cancellation or superposition characteristics of load and distributed energy fluctuations based on the correlation coefficient, and rationally regulate energy storage actions to reduce unnecessary energy storage. To reduce losses and improve energy storage regulation efficiency, the improved model predictive control framework used in the dynamic correction process constructs a time-domain variable-weight objective function based on a comprehensive fluctuation index. This automatically enhances the weighting of grid-connected power fluctuation suppression during periods of severe fluctuation, while prioritizing economic operation during periods of calm. Simultaneously, it limits the regulation amplitude of conventional units and maintains the healthy state of energy storage through fixed weights, achieving multi-objective rolling optimization of grid-connected power smoothing, unit regulation constraints, and energy storage health maintenance. The model predictive control framework incorporates a fluctuation component prediction feedback correction stage based on a long short-term memory neural network, which can proactively perceive the trend of fluctuation components in the future time domain and superimpose it as a correction term onto the initial scheduling plan prediction. The output curve enables advanced perception and pre-adjustment of future fluctuation uncertainties, improving the foresight and accuracy of scheduling. The entire automated scheduling method forms a complete closed loop from data acquisition, real-time collection, fluctuation decomposition, exponential calculation, threshold judgment to dynamic correction. It can respond to load and distributed energy fluctuations in real time, dynamically adjust the initial scheduling plan of conventional unit output and energy storage charging and discharging, and accurately issue correction commands. This effectively smooths system power fluctuations, improves the stability and adaptive adjustment capability of grid operation, optimizes the coordinated scheduling effect of conventional units and energy storage systems, reduces the frequent adjustment of conventional units, and balances the economy of grid operation with the durability of equipment operation. Attached Figure Description

[0044] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0045] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0046] like Figure 1 As shown, the method for automated scheduling based on load and distributed energy fluctuations includes the following steps:

[0047] Data acquisition: Acquire load forecast time-series data and distributed energy generation forecast time-series data, and generate an initial scheduling plan based on the load forecast time-series data and distributed energy generation forecast time-series data. The initial scheduling plan includes the output plan of conventional units and the charging and discharging plan of energy storage.

[0048] Data acquisition: Real-time acquisition of actual load power time-series data and distributed energy power time-series data;

[0049] Fluctuation decomposition: Adaptive variational mode decomposition is performed on the actual power time series data of load and the actual power time series data of distributed energy to extract load fluctuation components and distributed energy fluctuation components.

[0050] Index Calculation: Based on the load fluctuation component and the distributed energy fluctuation component, a comprehensive fluctuation index is obtained through a coupled fluctuation index calculation model. The coupled fluctuation index calculation model includes cross-coupling terms used to characterize the synchronicity of load and distributed energy fluctuations.

[0051] Threshold judgment: Determine whether the comprehensive fluctuation index exceeds the trigger threshold dynamically generated based on the real-time operating status. If it does, a dynamic correction process is triggered. Correction scheduling: During the dynamic correction process, based on the comprehensive fluctuation index and the current energy storage charge state, a multi-objective fuzzy decision-making mechanism is used to adjust the initial scheduling plan, generate a correction scheduling instruction, and issue it for execution.

[0052] The specific process of performing adaptive variational mode decomposition on the actual power time series data of load and the actual power time series data of distributed energy sources includes:

[0053] For the actual power time series data f of the load load (t) and time-series data of actual power of distributed energy sources f der (t) Construct variational constraint models respectively, and use the whale optimization algorithm to optimize the penalty factor α and the number of modal components K of variational mode decomposition;

[0054] The optimization process takes minimizing the envelope entropy as the objective function, calculates the envelope entropy value of each individual, and obtains the optimal parameter combination [αopt, Kopt] through iterative updates;

[0055] The original signal is decomposed based on optimal parameters to obtain a series of intrinsic mode functions (EMFs). The EMFs with center frequencies higher than the preset cutoff frequency are then reconstructed into load fluctuation components W. load (t) and distributed energy fluctuation component W der (t);

[0056] An adaptive variational mode decomposition (VMD) algorithm combining whale optimization is employed for the actual power time-series data of load and distributed energy sources. With the goal of minimizing envelope entropy, the penalty factor α and the number of mode components K in the VMD are optimized. This effectively avoids the signal over-decomposition or under-decomposition problems that easily occur when performing VMD with fixed α and K parameters in traditional methods. The resulting intrinsic mode functions (IMFs) better reflect the true signal characteristics of actual power fluctuations. The optimization objective of minimizing envelope entropy makes the envelope of the decomposed signal more regular and the fluctuation characteristics more prominent. Simultaneously, only IMFs with center frequencies higher than the preset cutoff frequency are reconstructed as fluctuation components, accurately separating the high-frequency fluctuation portion of the power time-series data and effectively eliminating interference from low-frequency trend terms. This accurately extracts load fluctuation components and distributed energy fluctuation components, providing high-precision, high-purity foundational data for subsequent calculation of the comprehensive fluctuation index, coupling analysis, and scheduling decisions, significantly improving the accuracy of subsequent fluctuation analysis and the reliability of scheduling decisions.

[0057] Select the actual load power time series data f of a certain 10kV distribution network load (t) and actual power time series data f of distributed energy (photovoltaic + wind power) der (t) is the object of analysis, and the sampling interval is set to 1 min, [85,72,90,68,82,65,78,62,80,58,75,60,79,63,83,59,77,61,81,57]kW.

[0058] Construct a variational constraint model: for f respectively load (t) and f der (t) Construct a variational constraint model for variational mode decomposition, the mathematical expression of which is: The constraints are ,in The set of decomposed intrinsic mode functions. The set of center frequencies of each intrinsic mode function. is the impulse function, * is the convolution operation, f(t) is the actual power time series data to be decomposed, j is the imaginary unit, and πt is the phase term.

[0059] Whale optimization algorithm settings: The whale optimization algorithm is introduced to optimize the penalty factor α and the number of modal components K in variational mode decomposition. The algorithm population size is set to 30, the maximum number of iterations is 50, the optimization range of the penalty factor α is [100, 500], and the optimization range of the number of modal components K is [2, 8]. The optimization objective function is to minimize the envelope entropy, and the formula for calculating the envelope entropy is: ,in , Let n be the envelope amplitude of the decomposed intrinsic mode function, and n be the number of sampling points for the envelope amplitude.

[0060] Load fluctuation component extraction: for f load (t) The whale optimization algorithm is executed to perform iterative optimization calculations, calculating the envelope entropy value of each individual in the population generation by generation. After 50 iterations, the optimal parameter combination is obtained when the envelope entropy E=0.82 reaches the global minimum. ;

[0061] Based on this optimal parameter, f load (t) Perform variational mode decomposition to obtain 5 eigenmode functions. The center frequencies of each intrinsic mode function, calculated through spectral analysis, are 0.13Hz, 0.09Hz, 0.06Hz, 0.03Hz, and 0.01Hz, respectively.

[0062] Set the preset cutoff frequency to 0.05Hz, and select centers with frequencies higher than this value. Reconstruction is performed to obtain the load fluctuation component. .

[0063] Extraction of distributed energy fluctuation components: for f der (t) Using the same whale optimization algorithm as described above, after 50 iterations, the optimal parameter combination is obtained when the envelope entropy E=0.79 reaches the global minimum. ;

[0064] Based on this optimal parameter, f der (t) Perform variational mode decomposition to obtain 4 eigenmode functions. The calculated center frequencies are 0.11Hz, 0.07Hz, 0.04Hz, and 0.02Hz, respectively.

[0065] Using 0.05Hz as the preset cutoff frequency, the center frequency is set to a value higher than this. Reconstruction is performed to obtain the distributed energy fluctuation components. .

[0066] The reconstructed and The authenticity was verified, and its fluctuation characteristics matched the real-time high-frequency fluctuations of load and distributed energy in the actual operation of the distribution network.

[0067] If traditional fixed parameters are used Variational mode decomposition was performed, but the reconstructed fluctuation components did not match the actual fluctuation situation well, and there was obvious over-decomposition. Some non-fluctuation trend terms were included in the fluctuation components. The comparison shows that the accuracy of the fluctuation components extracted in this case is much higher than that of the traditional fixed parameter decomposition method.

[0068] The specific process of obtaining the comprehensive volatility index through the coupled volatility index calculation model includes:

[0069] Within a preset sliding time window, Hilbert transforms are performed on the load fluctuation component and the distributed energy fluctuation component respectively to obtain the instantaneous amplitude function and instantaneous phase function at each moment within the time window;

[0070] The instantaneous amplitude ratio of the load fluctuation component and the distributed energy fluctuation component at each moment within the time window is calculated, and the instantaneous amplitude ratio at all moments within the time window is integrated to quantify the relative change in amplitude between the two, thus obtaining the first coupling coefficient C1.

[0071] The instantaneous phase difference between the load fluctuation component and the distributed energy fluctuation component at each moment within the time window is calculated, and the variance of the instantaneous phase difference at all moments within the time window is statistically analyzed to measure the synchronicity or misalignment of the two in the fluctuation time sequence, and the second coupling coefficient C2 is obtained.

[0072] The relative load fluctuation intensity is obtained by calculating the standard deviation of the load fluctuation component within the time window and dividing it by the base power value. Similarly, the relative distributed energy fluctuation intensity is obtained by calculating the standard deviation of the distributed energy fluctuation component within the time window and dividing it by the base power value. The square root of the sum of the squares of both components is then multiplied by an amplification factor consisting of a linear combination of the first and second coupling coefficients to obtain the comprehensive fluctuation index. The calculation formula is as follows:

[0073] ;

[0074] in, and These are the standard deviations of load and distributed energy fluctuation components, respectively. and The corresponding base power is γ1 and γ2 are preset coupling strength coefficients. By introducing cross-coupling terms, the superposition and amplification effect of the interaction between load and distributed energy fluctuations in amplitude and phase on the overall system volatility is characterized.

[0075] The coupled fluctuation index calculation model accurately captures the instantaneous amplitude and phase dynamic characteristics of load and distributed energy fluctuation components through Hilbert transform. It combines integral calculation and variance statistics to quantify the relative changes in amplitude and the synchronous misalignment characteristics of phase, respectively. The resulting coupling coefficient accurately characterizes the coupling relationship of fluctuations in amplitude and phase dimensions. Simultaneously, it calculates the comprehensive fluctuation index by combining the relative fluctuation intensity of load and distributed energy with the amplification factor formed by the coupling coefficient. The innovatively introduced cross-coupling term effectively characterizes the superimposed amplification effect of the interaction between load and distributed energy fluctuations in amplitude and phase on the overall system volatility. Compared to considering only the fluctuation intensity of load or distributed energy alone, this method can comprehensively, dynamically, and accurately quantify the overall system volatility level, avoiding the deficiency of a single fluctuation index in reflecting the superimposed effect of fluctuation coupling. It provides a scientific, accurate, and practical quantitative evaluation index for subsequent threshold judgment and dynamic scheduling correction, significantly improving the rationality and accuracy of subsequent scheduling decisions.

[0076] Based on the previous decomposition results, load fluctuation components obtained from adaptive variational mode decomposition of the same 10kV distribution network were selected. and distributed energy fluctuation components For the calculation object, the sampling interval is 1 minute, the sampling duration is 20 minutes, and there are a total of 20 sampling points. The specific data are as follows:

[0077] ;

[0078] ;

[0079] Setting parameters: Base load power (Taking the average actual power of the original load), Distributed Energy Base Power (Taking the average actual power of the original distributed energy resources), coupling strength coefficient The time window is 20 minutes (consistent with the sampling duration).

[0080] Hilbert transform obtains instantaneous features:

[0081] right and Perform Hilbert transforms on each component to obtain the instantaneous amplitude function of the load fluctuation component. Instantaneous phase function Instantaneous amplitude function of distributed energy fluctuation components Instantaneous phase function The instantaneous amplitude and phase of 20 sampling points were calculated (key feature selected: the integral interval of each instantaneous amplitude ratio is...). The instantaneous phase difference is measured in rad.

[0082] Calculate the first coupling coefficient C1 and the second coupling coefficient C2; First coupling coefficient C1: Calculate the instantaneous amplitude ratio at each moment. Perform integration within the time window [1,20], using the following formula: (T=20 is the time window length), the calculated amplitude ratios at each sampling point are 0.615, + ,0.333,3,0.9,20,2,9,1.250,+ ,2.000,7.5,1,4.333,0.364,14,2,5.333,0.667,20,After removing meaningless infinite values, the integral of the valid data is obtained. ,but ;

[0083] Second coupling coefficient C2: Calculates the instantaneous phase difference at each moment. Perform variance statistics on it, the formula is as follows: (n=20 is the number of sampling points, (The phase difference is the mean value). The phase differences of the 20 sampling points were calculated to be 0.21, 0.35, 0.18, 0.42, 0.25, 0.58, 0.32, 0.48, 0.28, 0.62, 0.30, 0.51, 0.24, 0.45, 0.19, 0.59, 0.31, 0.47, 0.22, and 0.60 rad, respectively. The mean value of the phase difference was then obtained. Substituting rad into the variance formula, we get C2 = 0.0189.

[0084] Calculate the relative fluctuation intensity of load and distributed energy resources:

[0085] Standard deviation of load fluctuation component: ,in Calculated ;

[0086] Standard deviation of distributed energy fluctuation components: , in Calculated ;

[0087] Relative load fluctuation intensity: ;

[0088] Relative volatility of distributed energy resources: .

[0089] Calculate the composite volatility index :

[0090] Substitute into the composite volatility index formula:

[0091] ;

[0092] Step 1: Calculate the part inside the square root:

[0093] ;

[0094] The second step is to calculate the magnification factor:

[0095] ;

[0096] The third step is to calculate the final index: .

[0097] Without introducing cross-coupling terms (i.e., amplification factor of 1), the calculated comprehensive fluctuation index is only 0.0751. This value does not reflect the superimposed amplification effect caused by the significant imbalance in amplitude (C1=4.315) and phase asynchrony (C2=0.0189) between load and distributed energy fluctuations, and deviates significantly from the degree of system fluctuation caused by fluctuation coupling in actual grid operation; while the value calculated in this case... It accurately reflects the true comprehensive fluctuation level under the coupling effect, providing a realistic quantitative basis for subsequent threshold judgment and avoiding scheduling decision deviations caused by exponential distortion.

[0098] The trigger threshold is a dynamic threshold generated based on the real-time running status. The specific acquisition process includes:

[0099] Obtain the current power grid frequency deviation And the state of charge (SOC) of energy storage, with the current grid frequency deviation Using the energy storage state of charge (SOC) as input variables, the frequency deviation adjustment factor k is directly output through a pre-generated two-dimensional fuzzy controller. f and the state of charge deviation adjustment factor k s ;

[0100] Based on the basic threshold T base Frequency deviation adjustment factor k f State of charge deviation adjustment factor k s Current power grid frequency deviation and the energy storage state of charge (SOC) and reference SOC ref Deviation | SOC-SOC ref Calculate the trigger threshold Specifically:

[0101] ;

[0102] Among them, SOC ref As a state of charge reference value, the threshold is adaptively changed according to the grid frequency health status and energy storage regulation capability;

[0103] The grid frequency deviation Δf and the energy storage state of charge (SOC) are used as input variables of the two-dimensional fuzzy controller, which dynamically outputs a frequency deviation adjustment factor k. f and the state of charge deviation adjustment factor k s By combining basic thresholds to construct a dynamic trigger threshold calculation model, the trigger threshold is made adaptively change with the grid frequency health status and energy storage regulation capability. This overcomes the shortcomings of traditional fixed thresholds that cannot adapt to the real-time operating status of the grid. When the grid frequency deviation is large and the energy storage regulation capability is insufficient, the threshold will be reasonably reduced to achieve early triggering of dynamic correction and timely smoothing of fluctuations to ensure grid stability. When the grid frequency is stable and the energy storage regulation capability is good, the threshold will be appropriately increased to avoid unnecessary frequent corrections and reduce ineffective regulation losses of conventional units and energy storage. This makes the triggering time of the dynamic correction process more in line with the actual operating needs of the grid, providing a scientific and dynamic judgment standard for the accurate triggering of subsequent dispatch corrections, and effectively improving the adaptability of automated dispatch and the stability of grid operation.

[0104] Based on the aforementioned 10kV distribution network scenario, the previously calculated comprehensive fluctuation index is used. The process revolves around calculating and determining the dynamic trigger threshold to ensure consistent scene parameters. The specific steps are as follows:

[0105] Set basic parameters:

[0106] Set the rated frequency of the power grid Energy storage state of charge reference value (Dimensionless, value range 0-1), the basic threshold for dynamic threshold calculation The two-dimensional fuzzy controller, after offline training and calibration, can accurately output the adjustment factor k based on the input Δf and SOC. f k s (The adjustment factor is a dimensionless coefficient, obtained by reasoning from the fuzzy rule base).

[0107] Obtain real-time running status parameters:

[0108] Real-time data collection of the power distribution network's operation yielded an actual network frequency of f = 49.92 Hz. The network frequency deviation Δf = ff was then calculated. N =49.92-50=-0.08Hz; Real-time monitoring of the energy storage system's state of charge yields the current actual value SOC=0.52.

[0109] Two-dimensional fuzzy controller output adjustment factor:

[0110] Inputting Δf=-0.08Hz and SOC=0.52 into a pre-generated two-dimensional fuzzy controller, and after fuzzification, rule inference, and defuzzification processes, the output frequency deviation adjustment factor k is determined. f=0.8, state of charge deviation adjustment factor k s =1.2.

[0111] Calculate the dynamic trigger threshold :

[0112] Substitute into the dynamic threshold calculation formula:

[0113] ;

[0114] Step-by-step calculation:

[0115] Calculate the absolute value of the state-of-charge deviation: ;

[0116] The part within the parentheses in the calculation formula is: 1 + 0.8 × (-0.08) + 1.2 × 0.08 = 1 - 0.064 + 0.096 = 1.032;

[0117] Calculate the final dynamic threshold: .

[0118] Threshold determination and effect verification:

[0119] This assessment will consider the overall volatility index. With dynamic threshold This triggers a dynamic correction process, allowing for timely adjustments to the initial scheduling plan;

[0120] Multi-condition comparison verification: If the power grid is operating well, the collected data are f=50Hz (Δf=0) and SOC=0.6 ( k is output by the fuzzy controller f =0, k s =0, calculated as follows Only when A correction is triggered when the value exceeds 0.12 to avoid frequent scheduling; if the power grid is in a vulnerable state, the collected values ​​are f=49.85Hz (Δ=-0.15Hz) and SOC=0.4 ( k is output by the fuzzy controller f =1.5, k s =1.8, calculated as follows The threshold is appropriately increased, and the correction trigger threshold is lowered when the energy storage regulation capacity is insufficient, so as to smooth out fluctuations in advance.

[0121] Comparison of drawbacks of fixed thresholds: If a fixed threshold T=0.12 is used, under the above-mentioned vulnerable power grid conditions, if... Under a fixed threshold, a correction will be triggered, but at this time the energy storage regulation capacity is insufficient. Early triggering (dynamic threshold 0.1362, if 0.13 is not exceeded, it will not be triggered temporarily, and the energy storage will be ready for regulation) is more in line with reality. In this case, the dynamic threshold can be flexibly adjusted according to the grid and energy storage status, making the trigger judgment more accurate.

[0122] The adjustment of the initial scheduling plan using a multi-objective fuzzy decision-making mechanism includes:

[0123] The comprehensive volatility index I cwi The state of charge (SOC) of energy storage and the correlation coefficient ρ between load fluctuation components and distributed energy fluctuation components are used as three-dimensional input variables into the multi-input fuzzy inference system.

[0124] Establish a fuzzy rule base with the dual objectives of minimizing system operating costs and maximizing power fluctuation mitigation, and output the power adjustment amount of conventional generating units. and energy storage power adjustment amount ;

[0125] The establishment of the fuzzy rule base is based on offline optimization. The parameters of the membership function are encoded and optimized through a genetic algorithm, so that the output of the fuzzy inference system approaches the Pareto optimal frontier in typical historical scenarios.

[0126] The comprehensive volatility index I cwi The State of Charge (SOC) of energy storage and the correlation coefficient ρ between load fluctuation components and distributed energy fluctuation components are introduced as three-dimensional input variables into the multi-input fuzzy inference system. A fuzzy rule base is constructed with the dual objectives of minimizing system operating costs and maximizing power fluctuation mitigation. Furthermore, a genetic algorithm is used to optimize the membership function parameters, bringing the rule base output close to the Pareto optimal front. The three-dimensional input comprehensively and accurately considers the overall system fluctuation level, the actual regulation capability of energy storage, and the coupling correlation characteristics of fluctuations, avoiding the one-sidedness of scheduling decisions caused by a single input dimension. The dual-objective rule base balances the economic efficiency of grid operation and... Stability is achieved by addressing the shortcomings of traditional single-objective optimization in scheduling. The optimization of the genetic algorithm makes the output of the fuzzy rule base in historical typical scenarios more in line with the optimal scheduling needs of the actual power grid. The multi-input fuzzy inference system can quickly infer and output the precise power adjustment of conventional units and energy storage, realize the coordinated optimization and regulation of conventional units and energy storage systems, reduce the frequent large-scale adjustment of conventional units and the ineffective charging and discharging losses of energy storage, improve the accuracy and real-time performance of scheduling instructions during dynamic correction, and make the adjustment of the initial scheduling plan more in line with the actual operating state of the power grid, effectively balancing the power fluctuation smoothing effect and system operating costs.

[0127] Based on the aforementioned unified scenario of a 10kV distribution network, the previously calculated comprehensive fluctuation index is used. With the energy storage state of charge (SOC) = 0.52, first calculate the Pearson correlation coefficient ρ between the load and the fluctuation components of distributed energy resources, and then infer the output power adjustment amount through a multi-objective fuzzy decision-making mechanism. The specific steps are as follows:

[0128] Determine the three-dimensional input variables

[0129] Given a fixed input: (dimensionless), SOC=0.52 (dimensionless, 0~1);

[0130] Calculate the Pearson correlation coefficient ρ: using the formula in this case, based on the previous load fluctuation components. and distributed energy fluctuation components;

[0131] The number of sampling points is n=20.

[0132] Calculate the mean. ,

[0133] ;

[0134] Calculate the numerator Calculate the product point by point and sum it:

[0135] ;

[0136] Calculate the denominator First, calculate the sum of squares of their respective deviations:

[0137] ,

[0138] Then calculate the product and take the square root: ;

[0139] Step 4: Calculate ρ. Since the Pearson correlation coefficient ranges from [-1, 1], we take ρ = -1.0 (indicating that the two are completely negatively correlated and their fluctuations cancel each other out).

[0140] The final three-dimensional input variables are: .

[0141] Multi-objective fuzzy reasoning system and rule base setting

[0142] System settings: This multi-input fuzzy inference system uses the Mamdani inference method, and the input domains are respectively... The output domain is the power adjustment of a conventional unit. Energy storage power adjustment amount ;

[0143] Rule base optimization: The fuzzy rule base aims to minimize system operating costs and maximize power fluctuation mitigation. It employs an offline genetic algorithm (population size 50, maximum iterations 100, crossover probability 0.8, mutation probability 0.05) to optimize the vertex parameter encoding of the triangular membership function, ensuring the output approaches the Pareto optimal frontier in typical historical scenarios. Core rules include: If... If SOC is medium and ρ is negative, then For small, To reduce the need for adjustment (by utilizing the mutual cancellation of fluctuations).

[0144] Fuzzy inference output power adjustment amount:

[0145] Three-dimensional input variables The input multi-input fuzzy inference system is fuzzified (mapping precise values ​​to fuzzy sets). Mapping to "medium", SOC mapping to "medium", ρ mapping to "negative large", rule matching and reasoning (matching the above core fuzzy rules), defuzzification (using the centroid method to convert the fuzzy output into a precise value), final output:

[0146] Power adjustment of conventional units (Slight increase in unit output);

[0147] Energy storage power adjustment (Energy storage involves small discharges; a negative sign indicates discharge, and a positive sign indicates charging.)

[0148] Based on the adjustment of the output, the initial scheduling plan was revised. Conventional units only increased their output by a small amount of 8kW, and energy storage only discharged a small amount of 5kW. By utilizing the completely negative correlation between load and distributed energy fluctuations (ρ=-1.0), the fluctuations were naturally canceled out, effectively smoothing out the overall system fluctuations. This also keeps the system operating cost low, with no significant regulation loss in conventional units, no ineffective charging and discharging of energy storage, and the energy storage SOC=0.52 is in a healthy range, so small discharges will not affect its regulation capability.

[0149] Single-objective scheduling comparison: If a scheduling method with the sole objective of maximizing power fluctuation mitigation is used, the output will be... While large-scale adjustments can quickly smooth out fluctuations, they incur additional losses from frequent large-scale adjustments by conventional units. Furthermore, large-scale discharges of energy storage cause a rapid decline in its state of charge, affecting subsequent regulation capabilities and significantly increasing system operating costs. If a scheduling method with minimizing operating costs as its sole objective is adopted, it will output... Without any adjustment, although there are no operating costs, the overall fluctuations of the system cannot be smoothed out, which will affect the stability of the power grid frequency and voltage.

[0150] Comparison of unoptimized rule bases: If the original fuzzy rule base, which has not been optimized by a genetic algorithm, is used, the output for the same input will be... The adjustment amount is between that of this case and the single-objective method, but its fluctuation smoothing effect is not as good as that of the single fluctuation smoothing objective, its operating cost is higher than that of this case, and its output does not approach the Pareto optimal frontier. In contrast, this case achieves the optimal balance of the two objectives after optimization by the genetic algorithm.

[0151] The correlation coefficient ρ is obtained by calculating the Pearson correlation coefficient between the load fluctuation component and the distributed energy fluctuation component. Specifically, it is calculated by multiplying the deviation of the load fluctuation component from its mean at each sampling time by the deviation of the distributed energy fluctuation component from its mean, and then dividing the sum of these products by the product of the standard deviations of the load fluctuation component and the distributed energy fluctuation component to obtain the correlation coefficient. This quantifies the degree of linear correlation between the two fluctuations. The calculation formula is as follows:

[0152] ;

[0153] Let be the load fluctuation component at time t. Let be the distributed energy fluctuation component at time t, and both are obtained by adaptive variational mode decomposition; This represents the average value of the load fluctuation component over the time window.

[0154] ρ is the mean of the distributed energy fluctuation component within the time window; n is the total number of sampling points within the time window; the correlation coefficient is used to determine whether the load and distributed energy fluctuations cancel each other out or overlap during the fuzzy inference process. When ρ is negative, the mutual cancellation between fluctuations is used to reduce energy storage action. When ρ is positive, the energy storage compensation is increased.

[0155] By calculating the Pearson correlation coefficient, the linear correlation between load and distributed energy fluctuation components can be accurately quantified. This clearly determines whether the fluctuations cancel each other out or overlap, providing a crucial and precise quantitative basis for multi-objective fuzzy decision-making mechanisms. When ρ is negative, it can guide dispatch to prioritize the mutual cancellation between fluctuations to reduce energy storage actions, significantly reducing energy storage charging and discharging losses and the frequency of ineffective regulation by conventional units. When ρ is positive, it can guide dispatch to increase energy storage compensation, effectively smoothing system power fluctuations after fluctuation overlap. This avoids the problems of excessive energy storage regulation or insufficient fluctuation smoothing caused by not considering the correlation characteristics of fluctuations. It makes the dispatch decisions in the fuzzy reasoning process more in line with the actual fluctuation coupling characteristics of the system, making them more targeted and scientific. This further improves the rationality of dynamic correction dispatch and the utilization efficiency of energy regulation resources, effectively balancing the stability and economy of grid operation.

[0156] Based on the aforementioned unified scenario of a 10kV distribution network, the load fluctuation component obtained through adaptive variational mode decomposition is used as before. and distributed energy fluctuation components The sampling interval is 1 minute, the sampling duration is 20 minutes, and the number of sampling points is n=20. All data must be kept consistent.

[0157] ,

[0158] ;

[0159] The specific steps are as follows:

[0160] The formula for calculating the Pearson correlation coefficient is as follows: The formula for calculating the Pearson correlation coefficient between load and the fluctuation component of distributed energy resources is:

[0161] ;

[0162] in This represents the average value of the load fluctuation component. The mean of the distributed energy fluctuation component is n=20, which is the number of sampling points.

[0163] Calculate the mean of the fluctuation components:

[0164] ;

[0165] ;

[0166] The numerator is the sum of the products of the deviations at each sampling time, i.e. After calculating the product of the deviations point by point and summing them, the final value is -526.15.

[0167] The denominator is the square root of the product of the sum of squares of the load fluctuation component deviation and the sum of squares of the distributed energy fluctuation component deviation, i.e. ;

[0168] Calculate the sum of squares of the load fluctuation component deviations:

[0169] ;

[0170] Calculate the sum of squared deviations of distributed energy fluctuation components: ;

[0171] Calculate the product and take its square root:

[0172] ;

[0173] Calculate the final Pearson correlation coefficient: ;

[0174] Since the Pearson correlation coefficient ranges from [-1, 1], we take ρ = -1. This value indicates that the load and the fluctuation of distributed energy are completely linearly negatively correlated, and the fluctuations of the two have a very strong mutual cancellation characteristic.

[0175] The application value of the coefficients can be illustrated by considering a scheduling scenario: After inputting ρ=-1 as one of the three-dimensional inputs into the fuzzy inference system of the multi-objective fuzzy decision-making mechanism, the system determines based on this value that the load and distributed energy fluctuations in the current power grid can naturally cancel each other out, without the need for significant energy storage regulation and conventional unit output adjustment. Therefore, it outputs a small amount of conventional unit power adjustment. Energy storage power adjustment amount The system fluctuations can be smoothed out with only a very small adjustment amount. Compared with the large adjustment scheme that does not take this coefficient into account, the energy storage charging and discharging loss is reduced by more than 90%, the adjustment frequency of conventional units is reduced, and the system operating cost is effectively saved.

[0176] If the operating conditions are changed, making the fluctuations of the two perfectly positively correlated (ρ=1, fluctuations superimposed), the fuzzy inference system will determine based on this value that the energy storage compensation needs to be increased, and output... The adjustment amount can quickly smooth out the large fluctuations after superposition, avoid large deviations in grid frequency and voltage, and ensure the stable operation of the grid.

[0177] If ρ=0 (the fluctuations of the two are not linearly correlated), the system will output a moderate adjustment amount, taking into account both fluctuation smoothing and operating costs, and achieving precise and adaptive scheduling decisions.

[0178] If the correlation between the two is not quantified using the Pearson correlation coefficient, the fuzzy inference system will be unable to accurately determine the canceling or superimposed characteristics of the fluctuations and will adopt a fixed adjustment strategy. In this example, under the condition of ρ=-1, it will still output a large adjustment amount, causing serious ineffective adjustment of energy storage and conventional units, and increasing the system operating cost. Under the superimposed condition of ρ=1, it will output a small adjustment amount, which cannot effectively smooth out the fluctuations and affect the stability of the power grid. However, this case, through precise quantitative judgment, enables the dispatch decision to achieve a precise match with the coupling characteristics of the fluctuations, which greatly improves the scientificity and rationality of the dispatch.

[0179] The dynamic correction process employs an improved model predictive control framework, constructing a time-domain variable-weight objective function using a comprehensive volatility index. The specific process includes:

[0180] Within the prediction time domain, the comprehensive volatility index is used as an adjustment factor for the weighting coefficients to construct a rolling optimization objective function, which consists of three parts:

[0181] The first part is the grid-connected power P. grid (k) and power reference value P refThe square of the difference, multiplied by a dynamic weight, is used to suppress grid-connected power fluctuations; the second part is the power adjustment amount for conventional units. The square of the product of the fixed weights The first part is used to limit the unit's regulation range; the third part consists of the energy storage state of charge (SOC(k)) and the state of charge reference value (SOC). ref The square of the difference multiplied by a fixed weight This is used to maintain the healthy state of energy storage, where SOC(k) is the SOC value of the energy storage at time k; the dynamic weight changes linearly with the comprehensive fluctuation index, and the calculation formula is as follows:

[0182] Among them, the weighting coefficient , With the base weight, β as the adjustment coefficient, and Np as the prediction time domain length, this system automatically enhances the weight for suppressing grid-connected power fluctuations when fluctuations are severe, while emphasizing economic operation when fluctuations are mild.

[0183] An improved model predictive control framework is adopted in the dynamic correction process. A time-domain variable-weight rolling optimization objective function is constructed using a comprehensive fluctuation index. The dynamic weights change linearly with the comprehensive fluctuation index, realizing adaptive regulation that automatically enhances the weight for suppressing grid-connected power fluctuations when fluctuations are severe and focuses on economic operation when fluctuations are mild. At the same time, the objective function integrates three core objectives: grid-connected power suppression, conventional unit adjustment range limitation, and maintenance of energy storage health status. By constraining excessive unit adjustment and energy storage state of charge deviation by fixed weights, the drawbacks of single-objective optimization are avoided. The rolling optimization method can continuously iterate and optimize the system state in the prediction time domain, making the scheduling correction more in line with the system fluctuation trend in the short term. It can effectively suppress grid-connected power fluctuations to ensure grid stability, reduce the mechanical losses of frequent and large-scale adjustments of conventional units, and maintain the energy storage state of charge in the healthy range to ensure its long-term regulation capability. Compared with traditional fixed-weight model predictive control, this method achieves a dynamic balance between grid operation stability and economy, and significantly improves the accuracy, adaptability, and foresight of dynamic correction scheduling.

[0184] Based on the aforementioned unified scenario of a 10kV distribution network, the previously calculated comprehensive fluctuation index is used. Energy storage state of charge (SOC) = 0.52, energy storage state of charge reference value. The calculations revolve around an improved model predictive control framework, with the core being the solution and verification of the variable-weight objective function. The specific steps are as follows:

[0185] Set the prediction time domain length (Unit: minutes, consistent with the sampling interval of 1 minute, covering the next 5 scheduling times k=1,2,3,4,5); Basic weights The comprehensive fluctuation index adjustment coefficient β=5; the adjustment range of conventional units has a fixed weight. Energy storage state of charge is maintained with a fixed weight. ; Grid-connected power reference value Pref = 200kW; Current time is t, and the grid-connected power at each time in the future prediction time domain is... Power adjustment of conventional units The energy storage state of charge (SOC(k)) is obtained through the initial scheduling plan and fluctuation prediction, and the specific value is as follows:

[0186] ;

[0187] ;

[0188] SOC(1)=0.53, SOC(2)=0.54, SOC(3)=0.55, SOC(4)=0.56, SOC(5)=0.57.

[0189] Calculate the dynamic weighting coefficients in the prediction time domain :

[0190] The formula for calculating dynamic weights is: Because the composite volatility index remains stable within the short-term forecast time domain, Since the values ​​are constant, the dynamic weights are consistent at all times, and the calculation is performed step by step:

[0191] ;

[0192] Calculate the rolling optimization objective function J:

[0193] The core formula of the objective function is:

[0194] ;

[0195] Substitute the parameters from k=1 to k=5 one by one to calculate the value of each term, and then sum them up. The process is as follows:

[0196] k=1:

[0197] ;

[0198] k=2:

[0199] ;

[0200] k=3:

[0201] ;

[0202] k=4:

[0203] ;

[0204] k=5:

[0205] ;

[0206] Summing yields the objective function value:

[0207] J=2507.6147+1155.6108+3880.1325+485.1298+1750.5027=9778.9905;

[0208] To demonstrate the adaptive advantages of dynamic weights, two extreme operating conditions are set up, keeping other parameters constant and only changing the comprehensive fluctuation index, and the dynamic weights and objective function values ​​are calculated:

[0209] Severely fluctuating operating conditions: (Far exceeding the dynamic threshold, system fluctuations are significant), calculated as follows Substituting these values ​​into the calculation yields J=14562.3585. The weight of the grid-connected power smoothing term in the objective function is significantly increased, and the system prioritizes suppressing grid-connected power fluctuations to ensure grid stability.

[0210] Smooth fluctuation operating conditions: (Below the dynamic threshold, the system runs smoothly), calculated as follows Substituting the values ​​into the calculation, we get J=7623.8852. The weight of the grid-connected power stabilization term is reduced, and the system focuses on controlling unit regulation and energy storage losses to improve operational economy.

[0211] If a traditional fixed-weight scheme is adopted, fixed weights are set. Substitute the original parameters to calculate the objective function value:

[0212] ;

[0213] Compared with the variable weight result J=9778.9905 in this case, the objective function value under fixed weight is lower, but its suppression of grid-connected power fluctuation is insufficient. According to the calculation, the average grid-connected power deviation under fixed weight is 7.2kW, while the average grid-connected power deviation under the variable weight method in this case is 4.8kW, and the fluctuation smoothing effect is improved.

[0214] If the fixed weight is increased to To match drastically fluctuating operating conditions, in this example Under normal operating conditions, the objective function value will rise to 14562.3585. Although this can smooth out fluctuations, it will increase the regulation losses of conventional units and the energy storage control costs, resulting in a significant decrease in economic efficiency.

[0215] The objective function in this case is... and By constraining the unit regulation and energy storage status, if the fixed weight constraint is removed and only the grid-connected power smoothing term is retained, the calculated J=7937.85, although the average grid-connected power deviation is reduced to 3.5kW, the adjustment amount of conventional units will increase to 15kW-20kW, resulting in large and frequent adjustments. The energy storage state of charge will deviate to 0.45-0.75, exceeding the healthy range, which seriously affects the subsequent regulation capability of energy storage. However, this method, through multi-objective weight constraints, smooths the fluctuations while controlling the unit adjustment amount within a small range of 6kW-9kW. The energy storage state of charge gradually approaches the reference value of 0.6, achieving a dual guarantee of stability and economy.

[0216] The model predictive control framework also includes a feedback correction stage based on fluctuation component prediction, the specific acquisition process of which includes:

[0217] The load fluctuation component W obtained from real-time decomposition load (t) and distributed energy fluctuation component W der (t) is input into a long short-term memory neural network to predict the trend of the fluctuation components in the future time domain;

[0218] The predicted fluctuation component trend is superimposed on the predicted output curve generated based on the initial scheduling plan, and used as a correction term for the prediction model in the model predictive control framework to achieve advance perception and pre-adjustment of future uncertainties.

[0219] By introducing a fluctuation component prediction feedback correction loop based on a long short-term memory neural network into the model predictive control framework, the long short-term memory neural network can accurately capture the temporal variation characteristics of load and distributed energy fluctuation components, effectively predict the trend of fluctuation components in the future time domain, and superimpose this predicted trend as a correction term onto the predicted output curve of the initial dispatch plan. This achieves advanced perception and pre-adjustment of future fluctuation uncertainties, making up for the shortcomings of traditional model predictive control that relies solely on historical and current data and is insufficient in predicting future fluctuations. This makes the predicted output curve more closely match the actual future grid operating conditions, significantly reducing the deviation of dispatch predictions, making the solution of the rolling optimization objective function more accurate, and thus making dispatch correction commands more forward-looking and targeted. It can pre-allocate the regulation resources of conventional units and energy storage, avoid large deviations in grid-connected power caused by sudden future fluctuations, reduce equipment losses and increased operating costs caused by temporary large-scale adjustments, further improve the optimization effect of the model predictive control framework and the accuracy of dispatch decisions, and enable the grid to have a stronger ability to respond to fluctuations in load and distributed energy, ensuring the stability of grid operation.

[0220] Based on the aforementioned unified scenario of a 10kV distribution network, and using the previously set prediction time domain length Np=5 (unit: minutes, corresponding to the next 5 scheduling times k=1,2,3,4,5), the grid-connected power reference value is... The load fluctuation components obtained by real-time decomposition Distributed energy fluctuation components ( Based on the data (consistent with the aforementioned), a long short-term memory neural network is used to predict and correct the fluctuation components. The specific steps are as follows:

[0221] LSTM Neural Network Model Setup and Training:

[0222] An LSTM model suitable for time series prediction of power fluctuation components is constructed. The model structure is "input layer - 2 LSTM hidden layers - fully connected output layer", with the input layer dimension being 2 (corresponding to...). The hidden layer has 64 neurons each, and the output layer has a dimension of 2 (corresponding to future time). The training set was selected from the time series data of load and distributed energy fluctuation components of the power distribution network over the past 30 days (sampling interval 1 min). The mean squared error (MSE) was used as the loss function, the optimizer was Adam, the learning rate was set to 0.001, and the training iterations were 200. After the model was trained, the prediction accuracy of the test set was higher, and it could accurately capture the time series change characteristics of the fluctuation components.

[0223] Input real-time fluctuation component data:

[0224] The load fluctuation components obtained from the real-time decomposition of 20 sampling points Distributed energy fluctuation components Using this as input to the LSTM model, the trend of the fluctuation components is predicted for the next Np=5 time points (k=1-5, corresponding to t=21-25min).

[0225] Predicting future fluctuation components in the time domain:

[0226] The trained LSTM model performs inference calculations to obtain the load fluctuation prediction components for the next five scheduling times. and distributed energy fluctuation prediction components The specific prediction results are as follows:

[0227] ;

[0228] ;

[0229] The prediction results provide accurate correction terms for the predicted output curve of the model predictive control.

[0230] Obtain the predicted output curve of the initial scheduling plan:

[0231] Based on the initial scheduling plan, the predicted output of conventional generating units for the next five time points is obtained. Predicted output of load foundation Distributed energy infrastructure predicts output The specific values ​​are:

[0232] ;

[0233] ;

[0234] ;

[0235] Based on the grid-connected power calculation formula, the initial uncorrected grid-connected power prediction value is obtained: The calculation yields:

[0236] .

[0237] The corrected predicted output curve is obtained by superimposing the correction term:

[0238] The fluctuation component predicted by LSTM is used as a correction term and superimposed onto the basic predicted output curve to obtain the corrected predicted load output. Distributed energy predictive output The predicted output of conventional generating units remains unchanged from its initial value (before dispatch correction), thus yielding the corrected grid-connected power prediction value: Step-by-step calculation:

[0239] k=1: ;

[0240] k=2: ;

[0241] k=3: ;

[0242] k=4: ;

[0243] k=5: ;

[0244] The corrected grid-connected power prediction curve incorporates advance predictions of future fluctuations, serving as the basis for solving the rolling optimization objective function in model predictive control.

[0245] After the power grid has been operating for the next 5 time periods (t=21-25min), the actual grid-connected power will be collected. Calculate the grid-connected power prediction deviations for 52kW, 75kW, 60kW, 82kW, and 55kW respectively, with and without feedback correction. The deviation formula is as follows: Then calculate the average deviation. :

[0246] Deviation without feedback correction: average deviation ;

[0247] Deviation with feedback correction: average deviation ;

[0248] Calculations show that after introducing the feedback correction stage, the average deviation of grid-connected power prediction decreased from 10.4kW to 1.6kW, improving prediction accuracy. Based on the corrected predicted output curve, the rolling optimization objective function is solved, and the generated scheduling correction instructions can adapt to future fluctuations in advance, ensuring that the grid-connected power at the next five time points approaches the reference value of 200kW after adjustment, without significant fluctuations. Without feedback correction, due to the large prediction deviation, the scheduling instructions cannot adapt to the actual fluctuations, resulting in multiple large deviations in grid-connected power, fluctuations in grid frequency and voltage, requiring temporary large-scale adjustments, and increasing losses of conventional units and energy storage.

[0249] If the traditional ARIMA time series forecasting method is used to predict future fluctuation components, the predicted results have a low degree of matching with the actual fluctuation components. The average deviation of the grid-connected power prediction after correction is 4.8kW, which is much higher than the 1.6kW of the LSTM model. This is because LSTM can effectively capture the nonlinear and long-term time series variation characteristics of power fluctuation components, while ARIMA is more suitable for linear stationary time series data. This further illustrates the accuracy of using LSTM for fluctuation component prediction in this method, making the optimization effect of the feedback correction link more significant.

[0250] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for automated scheduling based on load and distributed energy fluctuations, characterized in that, Includes the following steps: Data acquisition: Acquire load forecast time-series data and distributed energy generation forecast time-series data, and generate an initial scheduling plan based on the load forecast time-series data and distributed energy generation forecast time-series data. The initial scheduling plan includes the output plan of conventional units and the charging and discharging plan of energy storage. Data acquisition: Real-time acquisition of actual load power time-series data and distributed energy power time-series data; Fluctuation decomposition: Adaptive variational mode decomposition is performed on the actual power time series data of load and the actual power time series data of distributed energy to extract load fluctuation components and distributed energy fluctuation components. Index Calculation: Based on the load fluctuation component and the distributed energy fluctuation component, a comprehensive fluctuation index is obtained through a coupled fluctuation index calculation model. The coupled fluctuation index calculation model includes cross-coupling terms used to characterize the synchronicity of load and distributed energy fluctuations. The specific process includes: Within a preset sliding time window, Hilbert transforms are performed on the load fluctuation component and the distributed energy fluctuation component respectively to obtain the instantaneous amplitude function and instantaneous phase function at each moment within the time window; The instantaneous amplitude ratio of the load fluctuation component and the distributed energy fluctuation component at each moment within the time window is calculated, and the instantaneous amplitude ratio at all moments within the time window is integrated to quantify the relative change in amplitude between the two, thus obtaining the first coupling coefficient. The instantaneous phase difference between the load fluctuation component and the distributed energy fluctuation component at each moment within the time window is calculated, and the variance of the instantaneous phase difference at all moments within the time window is statistically analyzed to measure the synchronicity or misalignment of the two in the fluctuation time sequence, thus obtaining the second coupling coefficient. The standard deviation of the load fluctuation component within the time window is calculated and divided by its base power to obtain the relative load fluctuation intensity; the standard deviation of the distributed energy fluctuation component within the time window is calculated and divided by its base power to obtain the relative distributed energy fluctuation intensity; the square root of the sum of the squares of the two is multiplied by an amplification factor consisting of a linear combination of the first coupling coefficient and the second coupling coefficient to obtain the comprehensive fluctuation index. Threshold judgment: Determine whether the comprehensive volatility index exceeds the trigger threshold dynamically generated based on the real-time operating status. If it does, a dynamic correction process is triggered. Corrected scheduling: During the dynamic correction process, based on the comprehensive fluctuation index and the current energy storage state of charge, a multi-objective fuzzy decision-making mechanism is adopted to adjust the initial scheduling plan, generate corrected scheduling instructions, and issue them for execution; The dynamic correction process employs an improved model predictive control framework, constructing a time-domain variable-weight objective function based on a comprehensive volatility index. The specific process includes: Within the prediction time domain, the comprehensive volatility index is used as an adjustment factor for the weighting coefficients to construct a rolling optimization objective function, which consists of three parts: The first part is the square of the difference between the grid-connected power and the power reference value, multiplied by a dynamic weight, which is used to suppress grid-connected power fluctuations. The second part is the square of the conventional unit power adjustment amount multiplied by a fixed weight, which is used to limit the unit adjustment range; The third part is the square of the difference between the energy storage's state of charge and the reference value of the state of charge, multiplied by a fixed weight, used to maintain the health of the energy storage.

2. The method for automated scheduling based on load and distributed energy fluctuations according to claim 1, characterized in that: The specific process of performing adaptive variational mode decomposition on the actual power time series data of load and the actual power time series data of distributed energy sources includes: Variational constraint models are constructed for the actual power time series data of load and the actual power time series data of distributed energy, respectively. The penalty factor α and the number of modal components K of variational mode decomposition are optimized by introducing the whale optimization algorithm. The optimization process takes minimizing the envelope entropy as the objective function, calculates the envelope entropy value of each individual, and obtains the optimal parameter combination through iterative updates; The original signal is decomposed based on the optimal parameters to obtain several intrinsic mode functions. The intrinsic mode functions with a center frequency higher than the preset cutoff frequency are reconstructed into load fluctuation components and distributed energy fluctuation components.

3. The method for automated scheduling based on load and distributed energy fluctuations according to claim 2, characterized in that: The trigger threshold is a dynamic threshold generated based on the real-time running status. The specific acquisition process includes: The current grid frequency deviation and energy storage state of charge are obtained. Using the current grid frequency deviation and energy storage state of charge as input variables, the frequency deviation adjustment factor and the state of charge deviation adjustment factor are directly output through a pre-generated two-dimensional fuzzy controller. The trigger threshold is calculated based on the base threshold, frequency deviation adjustment factor, state of charge deviation adjustment factor, current grid frequency deviation, and deviation of energy storage state of charge from the reference value.

4. The method for automated scheduling based on load and distributed energy fluctuations according to claim 3, characterized in that: The adjustment of the initial scheduling plan using a multi-objective fuzzy decision-making mechanism includes: The comprehensive fluctuation index, energy storage state of charge, and the correlation coefficient between load fluctuation components and distributed energy fluctuation components are used as three-dimensional input variables and input into the multi-input fuzzy inference system. A fuzzy rule base is established with the dual objectives of minimizing system operating costs and maximizing power fluctuation mitigation effects. The outputs are the power adjustment amounts of conventional generating units and energy storage power adjustment amounts. The establishment of the fuzzy rule base is based on offline optimization. The parameters of the membership function are encoded and optimized through a genetic algorithm, so that the output of the fuzzy inference system approaches the Pareto optimal frontier in typical historical scenarios.

5. The method for automated scheduling based on load and distributed energy fluctuations according to claim 4, characterized in that: The correlation coefficient is obtained by calculating the Pearson correlation coefficient between the load fluctuation component and the distributed energy fluctuation component. Specifically, the deviation of the load fluctuation component from its mean at each sampling time is calculated and multiplied by the deviation of the distributed energy fluctuation component from its mean. The sum of these products is then divided by the product of the standard deviation of the load fluctuation component and the standard deviation of the distributed energy fluctuation component to obtain the correlation coefficient, thereby quantifying the degree of linear correlation between the two fluctuations. The correlation coefficient is used to determine whether load and distributed energy fluctuations cancel each other out or overlap during fuzzy inference. When the correlation coefficient is negative, the mutual cancellation between fluctuations is used to reduce energy storage action. When the correlation coefficient is positive, the energy storage compensation is increased.

6. The method for automated scheduling based on load and distributed energy fluctuations according to claim 5, characterized in that: The model predictive control framework also includes a feedback correction stage based on fluctuation component prediction. The specific acquisition process includes: The load fluctuation component and distributed energy fluctuation component obtained by real-time decomposition are input into a long short-term memory neural network to predict the trend of the load fluctuation component and distributed energy fluctuation component in the future time domain. The predicted fluctuation component trend is superimposed on the predicted output curve generated based on the initial scheduling plan, and used as a correction term for the prediction model in the model predictive control framework to achieve advance perception and pre-adjustment of future uncertainties.

Images