Energy storage capacity optimal configuration method and system based on flexible resource prediction

Abstract

The invention relates to an energy storage capacity optimal configuration method and system based on flexibility resource prediction, and the method comprises the steps: screening three modes of a base load, adjustability and impact of a precise stripping load through improving variational mode decomposition and Pearson correlation, and constructing an upper layer capacity-lower layer operation double-layer coupling frame; rapid global optimization is realized through an improved honeypot algorithm driven by a load mode, the operation is ensured to be feasible in a planning stage in cooperation with an adaptive penalty mechanism and a closed-loop verification system, and finally an energy storage capacity optimization configuration scheme considering the lowest cost of the whole life cycle and the efficient consumption of renewable energy sources is output; the method has the advantages that the double-layer coupling framework is constructed, the improved honeypot algorithm is adopted, the self-adaptive penalty function is constructed, multi-target dynamic balance is achieved, and a closed-loop verification system is established.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to a flexible resource prediction-based energy storage capacity optimization configuration method and system. BACKGROUND

[0002] With the in-depth development of intelligent manufacturing, the industrial field has higher requirements for the intelligentization of energy systems. As an important carrier for realizing efficient utilization of distributed renewable energy, power grids have been widely used in industrial parks. However, industrial loads have characteristics such as complex operation mode, severe power fluctuation, and highly dynamic production rhythm, which makes it difficult for traditional power grid design methods mainly based on residential or commercial scenarios to adapt to industrial applications. Under this background, building an optimization method that can identify the operation law of industrial loads, accurately match the output characteristics of renewable energy, and realize efficient collaborative configuration of source-load-storage systems has become a key to improving the operation efficiency and carbon reduction capacity of power grids.

[0003] The Chinese invention patent with the publication number CN120317427A proposes a sea star optimization algorithm-based energy storage system deployment optimization method and device, which is applied to the field of energy storage system deployment. The method includes establishing a multi-objective optimization function and constraint conditions of the energy storage system according to the optimization target. Under the constraint conditions, the sea star optimization algorithm is used to solve the multi-objective optimization function of the energy storage system. The position of the sea star individual with the lowest fitness in all sea star individuals is obtained when the preset termination iteration condition is reached by iterating the position and fitness of the sea star individual. The position of the sea star individual with the lowest fitness is taken as the final deployment scheme of the energy storage system. In this way, the deployment performance and optimization efficiency of the energy storage system are significantly improved, the service life of the energy storage system equipment is prolonged, the minimization of cost, the maximization of system reliability and energy utilization rate are realized, and the optimization process is more flexible, practical and applicable.

[0004] The above-mentioned technology still needs to be further solved in the following problems:

[0005] 1. The traditional Fourier or wavelet decomposition cannot distinguish between transient impacts and steady-state baselines in industrial loads, resulting in overestimation of energy storage capacity or power shortage at critical moments.

[0006] 2. The static configuration model treats the operation strategy as a "black box", and the planning results are disconnected from the actual scheduling, often resulting in a gap between the design consumption rate of 90% and the operation rate of only 70%.

[0007] 3. Traditional algorithms such as genetic and particle swarm are randomly initialized and have fixed steps, which are prone to premature convergence in load multi-peak surfaces, resulting in large differences in repeated operation results and difficulty in engineering solidification.

[0008] 4. The fixed weight penalty function ignores changes in load structure, and the penalty for curtailment of adjustable loads is insufficient, while the penalty for carbon emissions in periods dominated by base load is excessive.

[0009] 5. Current verification only compares static investment costs, without measuring energy storage cycle efficiency or curtailment rate. Only after commissioning are large-scale failures of design indicators discovered, resulting in high rectification costs. Therefore, it is necessary to provide a method and system for optimizing energy storage capacity configuration based on flexible resource prediction, which constructs a two-layer coupling framework, adopts an improved honeypot algorithm, constructs an adaptive penalty function, achieves multi-objective dynamic balance, and establishes a closed-loop verification system. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for optimizing energy storage capacity based on flexible resource prediction, which constructs a two-layer coupling framework, adopts an improved honeypot algorithm, constructs an adaptive penalty function, achieves multi-objective dynamic balance, and establishes a closed-loop verification system.

[0011] The objective of this invention is achieved as follows: a method for optimizing energy storage capacity allocation based on flexible resource prediction, the method comprising the following steps:

[0012] S1: Energy storage resource data acquisition based on flexible resources: Real-time data acquisition is carried out for various production equipment, auxiliary facilities and public loads in industrial plants through smart meters deployed at key nodes of the power distribution network and the plant energy management system.

[0013] S2: Multimodal feature decomposition and clustering of energy storage resource data: By improving variational mode decomposition and Pearson correlation screening, the three modes of load, namely base load, adjustable load, and impact load, are accurately separated;

[0014] S3: Construction of Source-Load-Storage Coordinated Optimization Model: Construct a two-layer coupled framework of "upper-layer capacity-lower-layer operation" and achieve coupled optimization of source-load-storage parameters through two-layer coordinated iteration;

[0015] S4: Improved honeypot algorithm for optimization: The improved honeypot algorithm based on load pattern category-driven algorithm achieves fast global optimization, and is combined with an adaptive penalty mechanism and closed-loop verification system to ensure operational feasibility in the planning stage;

[0016] S5: Grid-storage collaborative optimization configuration: Based on the aforementioned optimization results, output an energy storage capacity optimization configuration scheme that balances the lowest life-cycle cost and efficient renewable energy consumption.

[0017] The data acquisition of energy storage resources based on flexible resources in S1 includes the following specific acquisition process:

[0018] Data collection coverage: encompasses all power-consuming units within the factory area, including main equipment of continuously operating production lines, auxiliary equipment that starts and stops intermittently, and sudden impact loads;

[0019] Collection parameter setting: take 1 minute as the sampling interval, continuously collect at least one complete production cycle of active power time series data, and synchronously record the time stamp, device running state and working condition label;

[0020] Data quality assurance: cross-check abnormal values through redundant metering devices, and use the plant communication network to transmit data to the central database in real time to form a standardized format of raw load power data set, providing high-precision input for subsequent multi-modal feature decomposition.

[0021] The multi-modal feature decomposition and clustering of the energy storage resource data in S2 specifically includes the following steps:

[0022] S2.1: Variational modal decomposition model construction: build an optimization model to ensure that the sum of all modal components is equal to the original energy storage resource data by minimizing the time derivative constraint of the modal component, so as to decompose the sub-load mode component;

[0023] S2.2: Modal component screening: calculate the Pearson correlation coefficient of each modal component and the original energy storage resource data, and retain the components with a correlation coefficient exceeding a preset threshold, to ensure that the selected modal components are highly correlated with the original load;

[0024] S2.3: Typical load mode extraction: classify the screened modal components into base load mode components, adjustable mode components and impact mode components, so as to identify and extract three typical load mode characteristics.

[0025] The source-load-energy storage collaborative optimization model in S3 is constructed, specifically including the following steps:

[0026] S3.1: Upper device configuration optimization: optimize the device configuration decision vector, including photovoltaic installed capacity, wind turbine installed capacity, energy storage system capacity and rated power, to minimize the sum of device investment cost and expected operation cost, and determine the optimal device configuration parameters;

[0027] S3.2: Lower operation strategy optimization: under the given device configuration, optimize the operation strategy vector to minimize the total operation cost of grid power purchase cost, abandoned load penalty and energy storage cycle loss penalty, and determine the optimal scheduling scheme.

[0028] The improved honeypot algorithm in S4 is optimized, specifically including the following steps:

[0029] S4.1: Improved honeypot algorithm reconnaissance bee initialization: based on the load mode category, use the historical mean and standard deviation of photovoltaic and wind turbine output, energy storage proportion coefficient, maximum daily power consumption and maximum power under this mode, combined with random disturbance to generate initial honeypot position;

[0030] S4.2: Adaptive neighborhood search in the following bee phase: an adaptive mechanism driven by load fluctuation is adopted, first, the load fluctuation weight of each period is calculated, then the adaptive step factor is generated by combining the fitness of the honeypot, and finally the position of the honeypot is updated by the fluctuation weight weighting;

[0031] S4.3: Dynamic disturbance mechanism in the observation bee phase: a directional disturbance mechanism is adopted, first, the similarity between the current load and the historical mode is calculated, then the random disturbance intensity is generated, and the similarity weighted directional disturbance is applied to the position of the honeypot, and the local optimal escape is realized by combining the global optimal information;

[0032] S4.4: Penalty function calculation of multi-objective constraint: an adaptive penalty mechanism associated with load characteristics is constructed, the renewable energy consumption penalty weight is dynamically adjusted based on the adjustable load total power proportion, and the carbon emission penalty weight is adjusted based on the average power proportion of the base load, then the penalty term is added to the total cost to reconstruct the fitness function, realizing the multi-objective collaborative optimization;

[0033] S4.5: Configuration scheme generation and verification: multi-dimensional verification is carried out through typical day scene simulation, first, the cost deviation rate is calculated to ensure the prediction accuracy, then the curtailment rate is tested to meet the renewable energy utilization requirement, then the energy storage cycle efficiency is verified to meet the performance standard, and the robustness of the scheme in actual operation is guaranteed.

[0034] The adaptive neighborhood search in the following bee phase in S4.2 is as follows:

[0035] S4.21: Fluctuation rate weight calculation: the load fluctuation weight of each period is calculated, based on the ratio of the standard deviation of the load in this period to the maximum standard deviation of all periods, the fluctuation intensity of different periods is quantified;

[0036] S4.22: Adaptive step factor calculation: based on the ratio of the honeypot fitness value to the current maximum fitness value, combined with the reference step coefficient, the adaptive step factor of each dimension is calculated;

[0037] S4.23: Honeypot position update: update the position of the honeypot, through the current value plus the adaptive step factor multiplied by the difference with the random honeypot, combined with the fluctuation weight, realize the load fluctuation weighted neighborhood search operation.

[0038] The dynamic disturbance mechanism in the observation bee phase in S4.3 is as follows:

[0039] S4.31: Load pattern similarity calculation: calculate the Euclidean distance between the current load curve and each historical mode curve, and convert it into a similarity value;

[0040] S4.32: Disturbance intensity generation: randomly generate disturbance intensity coefficient, uniformly distributed in the preset interval, control the disturbance size to enhance the flexibility of the algorithm;

[0041] S4.33: Honey pot position disturbance: Apply disturbance to the honey pot position, the disturbance amount is the product of the disturbance intensity and the difference between the global optimal honey pot and the current honey pot, and the similarity is combined to realize similarity weighted directional disturbance to escape from the local optimum.

[0042] The penalty function calculation of the multi-objective constraint in S4.4 is as follows:

[0043] S4.41: Constraint penalty term calculation: Calculate the constraint penalty term, weight the total power proportion of adjustable load based on the degree of renewable energy consumption rate deficiency, weight the average power proportion of base load based on the degree of carbon emission exceeding, and dynamically adjust the penalty weight to handle multi-objective constraints;

[0044] S4.42: Fitness function reconstruction: Add the constraint penalty term to the original total cost, adjust through the penalty coefficient, and reconstruct the fitness function to comprehensively consider cost optimization and constraint satisfaction.

[0045] The configuration scheme generation and verification in S4.5 are as follows:

[0046] S4.51: Optimal configuration scheme output: Output the global optimal configuration scheme vector after the convergence of the honey pot optimization algorithm as the final configuration result;

[0047] S4.52: Cost deviation rate verification: Calculate the relative deviation between the total cost predicted by the optimization algorithm and the running cost based on the typical day scenario simulation, and ensure that the deviation is lower than the preset threshold to verify the cost accuracy;

[0048] S4.53: Abandonment rate verification: Calculate the proportion of the total renewable energy abandonment power and the total actual output, and ensure that the abandonment rate does not exceed the preset threshold to verify the utilization efficiency of renewable energy;

[0049] S4.54: Energy storage cycle efficiency verification: Calculate the ratio of the total energy storage discharge energy and the total charging energy as the cycle efficiency, and ensure that the efficiency is higher than the minimum required threshold to verify the performance of the energy storage.

[0050] The grid energy storage collaborative optimization configuration in S5 is as follows:

[0051] S5.1: Configuration scheme deployment: Convert the generated optimal configuration scheme vector into entity device selection;

[0052] S5.2: Collaborative operation strategy embedding: Write the optimal operation strategy of the lower layer optimization into the grid energy management system.

[0053] The beneficial effects of the present application are:

[0054] 1. Combine "variational modal decomposition + Pearson correlation coefficient screening" with industrial load analysis to accurately split strong fluctuation load into base load, adjustable load and impact load, and solve the problem of transient characteristics being easily submerged.

[0055] 2. Propose a "upper-layer capacity configuration - lower-layer operation strategy" double-loop framework to put planning variables and operation variables in the same iteration channel to realize the true coupling optimization of source-load-storage parameters.

[0056] 3. Design a load mode driven improved honeypot algorithm to generate initial honeypot of reconnaissance bees with mode statistics, use period fluctuation rate to weight step length, and use historical similarity to trigger directional disturbance, so that the search is fast and stable.

[0057] 4. Construct an adaptive penalty function of "load characteristics - constraint weight", which automatically increases the penalty for curtailment when the proportion of adjustable load is high, and strengthens the carbon emission constraint when the proportion of base load is high, to achieve dynamic balance of multiple objectives.

[0058] 5. Establish a "configuration-simulation-feedback" closed-loop verification system to examine cost deviation, curtailment rate and energy storage cycle efficiency under typical day scenarios, and ensure that the optimization scheme still has robustness in real fluctuation environment. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 The flowchart of the present application.

[0060] Figure 2 The load characteristic decomposition result diagram of the present application.

[0061] Figure 3 The convergence process comparison diagram of the optimization algorithm of the present application.

[0062] Figure 4 The optimization result stability comparison (50 runs) diagram of the present application.

[0063] Figure 5 The optimization algorithm calculation efficiency comparison diagram of the present application.

[0064] Figure 6 The renewable energy utilization rate comparison diagram of the optimization scheme of the present application.

[0065] Figure 7 The cost comparison diagram of different configuration schemes of the present application.

[0066] Figure 8 The system structure block diagram of the present application. DETAILED DESCRIPTION

[0067] The present application will be further described below in conjunction with the embodiments and / or drawings.

[0068] Embodiment 1

[0069] As Figures 1-7 shown, a flexible resource prediction-based energy storage capacity optimization configuration method, the method comprises the following steps:

[0070] S1: Energy storage resource data collection based on flexible resources: for various production equipment, auxiliary facilities and public loads in the industrial plant, real-time data collection is carried out through the intelligent electric meter deployed at the key nodes of the power distribution network and the plant energy management system (SCADA);

[0071] In the present application, the specific collection process comprises:

[0072] Collection object coverage: covers all power consumption units in the plant, including continuously running production line main equipment (such as compressors, pump groups), intermittently starting and stopping auxiliary equipment (such as cooling tower fans, conveyors), and sudden impact loads (such as large motor starting, arc furnace switching);

[0073] Collection parameter setting: with 1 minute as the sampling interval, continuously collect active power time series data for at least one complete production cycle (usually 1 year), and synchronously record time stamp, device running state (start / stop / standby) and working condition label (such as production batch, shift information);

[0074] Data quality assurance: eliminate abnormal values through cross verification of redundant metering devices, and use the plant communication network to transmit data to the central database in real time to form a standardized format original load power data set (denoted as L t , t is the time index), providing high-precision input for subsequent multi-modal feature decomposition.

[0075] S2: Multi-modal feature decomposition and clustering of energy storage resource data: through improved variational modal decomposition and Pearson correlation screening, the base load, adjustable load and impact load of the load are accurately stripped;

[0076] In the present application, the energy storage resource data has strong volatility, multi-time scale characteristics and operating mode switching characteristics, and it is difficult to accurately separate transient impact and steady-state mode using conventional Fourier transform or wavelet decomposition method, which easily causes modal aliasing and feature loss problems, resulting in inaccurate load characteristic extraction.

[0077] In the present application, through the improved variational modal decomposition combined with the Pearson correlation coefficient screening mechanism, the sub-load mode components are first decomposed by constructing an optimization model to avoid modal aliasing, then the correlation coefficients of each modal component and the original load are calculated, the high correlation components are retained to solve the feature loss problem, and finally the screened components are classified into base load mode, adjustable mode and impact mode to extract typical load characteristics.

[0078] Specifically comprising the following steps: ① variational modal decomposition model construction

[0079] An optimization model is constructed, the sum of all modal components is ensured to be equal to the original energy storage resource data by minimizing the time derivative constraint of the modal component, so as to decompose the sub-load mode component, effectively extract the typical load mode and avoid modal aliasing;

[0080] The optimization objective function is: The constraint condition is: In the formula, u m (t) is the mth modal component, representing the decomposed sub-load mode, m = 1, 2,..., M; ω m is the center frequency corresponding to u m (t); is the objective function minimized with respect to the modal component {u m} and the center frequency {ω m}; M is a preset modal number, such as M = 6; denotes the partial derivative with respect to time t; δ(t) is the Dirac function; * is the convolution operator; j is the imaginary unit; is the L2 norm; L t is the original load power input data at time t.

[0081] ② Modal component screening

[0082] The Pearson correlation coefficient of each modal component and the original energy storage resource data is calculated, the components with a correlation coefficient exceeding a preset threshold are retained, and the screened modal components are highly correlated with the original load, solving the feature loss problem, and is expressed as: In the formula, ρ m is the Pearson correlation coefficient of the mth modal component u m and the original load L t ; cov(·,·) represents the covariance operation; is the standard deviation of u m (t); is the standard deviation of the original load L t ; θ ρ is a correlation coefficient screening threshold, such as θ ρ = 0.7.

[0083] ③ Typical load mode extraction

[0084] The screened modal components are classified into base mode components, adjustable mode components and impact mode components, so as to identify and extract three typical load mode characteristics;

[0085] The base load mode component u1(t) represents a stable continuous load; the adjustable mode component u2(t) represents an adjustable load fluctuation; and the impact mode component u3(t) represents a burst transient load.

[0086] As a specific embodiment that can be implemented, a multimodal feature decomposition effect comparison analysis is carried out to verify the superiority of the improved variational modal decomposition combined with the Pearson correlation coefficient screening mechanism in industrial load feature extraction; as shown in the figure, Figure 2 The decomposition accuracy of three typical load modes (base load mode, adjustable mode and impact mode) is investigated by comparing the traditional Fourier transform, wavelet decomposition, traditional variational modal decomposition and the method of the application; Figure 2 The upper part shows the decomposition effect of the method of the application on the energy storage resource data: the original load curve is accurately separated into a stable base load mode, a regular fluctuation adjustable mode and a burst impact mode, and the boundaries of the three are clear without overlapping; the middle part shows the traditional variational modal decomposition result, and the base load mode is obviously mixed into the adjustable fluctuation, and the impact load is dispersed into multiple modes, and the modal aliasing problem occurs; the lower part of the column chart quantitatively compares the feature extraction accuracy, and the accuracy of the method of the application on the three modes is significantly higher than that of the traditional method, and the accuracy of the impact mode is improved most obviously.

[0087] The experimental results fully prove that the method of the application effectively solves the modal aliasing and feature loss problems caused by the strong fluctuation of the industrial load by optimizing the objective function construction and the correlation coefficient screening mechanism.

[0088] S3: Source-load-energy storage collaborative optimization model construction: a double-layer coupled framework of "upper layer capacity-lower layer operation" is constructed, and the coupling optimization of source-load-energy storage parameters is realized through two-layer collaborative iteration;

[0089] In the application, due to the uncertainty of photovoltaic and wind turbine output and the multi-mode characteristics of load, the conventional configuration model ignores the coupling of device parameters and operation strategy, resulting in that the planning result does not match the actual operation scene, and the "planning-operation decoupling" problem cannot be solved.

[0090] The application establishes a double-layer optimization framework, the upper layer optimizes the device configuration parameters to minimize the sum of device investment cost and expected operation cost, and the lower layer optimizes the operation strategy under the given configuration to minimize the grid power purchase cost, load curtailment penalty and energy storage loss cost, and the coupling optimization of source-load-energy storage parameters is realized through two-layer collaborative iteration.

[0091] The specific steps are as follows: ① upper layer device configuration optimization

[0092] The optimization device configuration decision vector includes photovoltaic installed capacity, wind turbine installed capacity, energy storage system capacity and rated power, and the sum of device investment cost and expected operation cost is minimized to determine the optimal device configuration parameters, which is represented as: where X is the configuration decision vector, denoted as denotes the installed capacity of photovoltaic power; denotes the installed capacity of wind turbine; denotes the installed capacity of energy storage system; denotes the rated power of energy storage system; C inv (X) is the equipment investment cost function; E[·] denotes the expected operation cost operator; C op (X, S * (X)) denotes the optimal operation strategy S * corresponding to the operation cost under configuration X; S * (X) is the optimal operation strategy output by the lower-level optimization.

[0093] ② Lower-level operation strategy optimization

[0094] Under a given equipment configuration, the optimal operation strategy vector is optimized to determine the optimal scheduling scheme by minimizing the total operation cost of grid electricity purchase cost, load shedding penalty and energy storage cycle loss penalty, denoted as: where S is the operation strategy vector; T is the total number of optimization periods; is the operation cost function C op minimized with respect to the operation strategy S; c grid (P grid,t ) is the cost function of grid electricity purchase at t period; P grid,t is the grid electricity purchase power at t period; ΔL curt,t is the load shedding amount at t period; λ curt is the load shedding penalty coefficient; is the energy storage charging and discharging power at t period, charging is positive and discharging is negative; is the energy storage cycle loss penalty coefficient; |·| denotes the absolute value operation.

[0095] S4: Improved honeypot algorithm for optimization: improved honeypot algorithm based on load mode category driving realizes fast global optimization, and cooperates with adaptive penalty mechanism and closed-loop verification system to ensure operation feasibility in the planning stage;

[0096] Specifically, the following steps are included: ① Improved honeypot algorithm reconnaissance bee initialization

[0097] The traditional honeypot optimization algorithm randomly generates initial solution, which is easy to fall into local optimum and has low search efficiency, and does not utilize load mode characteristic information, resulting in slow convergence speed and unstable optimization quality.

[0098] This invention is based on load mode categories. It utilizes the historical average and standard deviation of photovoltaic and wind turbine output, energy storage ratio coefficient, maximum daily power consumption and maximum power of the load under these modes, and combines random perturbations to generate initial honeypot locations so that the search can be initiated in the high-quality solution region. The specific steps are as follows: In the formula, Let k be the location of the i-th initial honeypot; k is the load pattern category index. This represents the historical average of photovoltaic power output under the k-th load type. This represents the historical standard deviation of photovoltaic power output under the k-th load mode; This represents the historical average output of the fan under the k-th load mode; ξ represents the historical standard deviation of the wind turbine output under the k-th load mode; ξ is a standard normally distributed random variable, i.e., ξ ~ N(0,1); This represents the maximum daily power consumption of the k-th type of load. α is the maximum power of the k-th type of load; k β is the proportional gain of energy storage capacity, with a value range of [0.2, 0.4]; k is the energy storage power proportionality coefficient, with a value range of [0.3, 0.5]; [·,·,·,·] is the vector construction symbol, representing a column vector containing four elements.

[0099] ② Adaptive neighborhood search during the bee-following stage

[0100] Traditional algorithms use a fixed search step size, which is prone to premature convergence in multi-peak industrial load optimization scenarios. They cannot adapt to the spatiotemporal differences in load fluctuations, and it is especially difficult to maintain search diversity during critical periods.

[0101] This invention employs a load fluctuation-driven adaptive mechanism. First, it calculates the load fluctuation weight for each time period. Then, it combines the honeypot fitness to generate an adaptive step size factor. Finally, it updates the honeypot position using a weighted average of the fluctuation weights, achieving refined searching during critical time periods. The specific steps are as follows:

[0102] 1) Volatility weighting calculation

[0103] Calculate the load fluctuation weight for each time period. Based on the ratio of the load standard deviation for that time period to the maximum standard deviation across all time periods, quantify the fluctuation intensity of different time periods to guide subsequent searches, expressed as: In the formula, η j Let be the volatility weight for the j-th time period; The standard deviation of historical load power in time period j represents the intensity of fluctuations during that period; max(σ L ) represents the maximum value of the standard deviation of load across all time periods; j is the time period index, such as j=1 corresponding to the valley period; j=2 corresponding to the peak period, etc.

[0104] 2) Adaptive step size factor calculation

[0105] Based on the ratio of the honeypot fitness value to the current maximum fitness value, combined with the baseline step size coefficient, an adaptive step size factor is calculated for each dimension to dynamically adjust the search range to adapt to optimization requirements, expressed as: In the formula, φ i,j f(X) is the adaptive step size factor of the i-th honeypot in the j-th dimension; i ) represents the fitness value of the i-th honeypot; ω represents the maximum fitness value among all current honeypots; φ The reference step size coefficient, such as ω φ =0.5.

[0106] 3) Honeypot location update

[0107] Update the honeypot location by adding an adaptive step size factor to the current value and multiplying it by the difference from the random honeypot, then combining this with fluctuation weights to achieve a load fluctuation-weighted neighborhood search operation, represented as: In the formula, This is the updated j-th dimension parameter value for the i-th honeypot; This represents the j-th dimension parameter value of the current i-th honeypot; Let r be the j-th dimension parameter value for randomly selected honeypot r; r is the index of the random honeypot, r≠i.

[0108] ③ Observe the dynamic disturbance mechanism during the bee stage.

[0109] Traditional methods struggle to effectively escape local optima, especially when load patterns change abruptly, lacking a rapid response mechanism, resulting in optimization results that cannot adapt to dynamic changes.

[0110] This invention employs a directional perturbation mechanism. First, it calculates the similarity between the current load and historical patterns. Then, it generates a random perturbation intensity and applies a similarity-weighted directional perturbation to the honeypot location. Combining global optimal information, it achieves local optimal escape, as detailed below:

[0111] 1) Load pattern similarity calculation

[0112] Calculate the Euclidean distance between the current load curve and each historical pattern curve, convert it into a similarity value, and the smaller the distance, the higher the similarity, to measure the degree of matching between the current scene and the historical patterns, expressed as: In the formula, sim k L represents the similarity between the current load curve and the k-th historical pattern. current This represents the load power curve vector at the current moment; Let ||k|| represent the typical curve vector of the k-th type of historical load pattern; ||·||2 represents the norm of the vector.

[0113] 2) Generation of disturbance intensity

[0114] Randomly generate disturbance intensity coefficient γ, uniformly distributed in a preset interval, control the disturbance size to enhance the flexibility of the algorithm, and the preset interval adopts uniform distribution U(0, 0.3), that is, U(0, 0.3) represents uniform distribution in the interval [0, 0.3], and 0.3 is the upper limit of the disturbance intensity.

[0115] 3) Honey pot position disturbance

[0116] The disturbance is applied to the position of the honey pot, the disturbance amount is the disturbance intensity multiplied by the difference between the global optimal honey pot and the current honey pot, and the similarity is combined to realize the similarity weighted directional disturbance to escape from the local optimum, and is represented as: In the formula, is the position vector of the i-th honey pot after disturbance; X i is the position vector of the current i-th honey pot; X best is the position vector of the current global optimal honey pot.

[0117] (4) Penalty function calculation of multi-objective constraint

[0118] The conventional penalty function method adopts fixed weight for renewable energy consumption rate and carbon emission constraint, and does not consider the dynamic influence of load characteristics, so that the consumption constraint is insufficient in the period dominated by adjustable load, and the carbon constraint is insufficient in the period dominated by base load.

[0119] The application constructs an adaptive penalty mechanism related to load characteristics, dynamically adjusts the renewable energy consumption penalty weight based on the total power proportion of adjustable load, and adjusts the carbon emission penalty weight based on the average power proportion of base load, then adds the penalty term to the total cost to reconstruct the fitness function, realizes multi-objective collaborative optimization, and the specific process is as follows:

[0120] 1) Constraint penalty term calculation

[0121] The constraint penalty term is calculated, the total power proportion of adjustable load is weighted based on the degree of renewable energy consumption rate deficiency, and the average power proportion of base load is weighted based on the degree of carbon emission exceeding, and the penalty weight is dynamically adjusted to process multi-objective constraint, and is represented as: In the formula, Ψ(X) is the constraint penalty term; is the minimum renewable energy consumption rate requirement; is the actual renewable energy consumption rate; is the maximum allowed carbon emission amount; C em is the actual carbon emission amount; E L_flex represents the total power of adjustable load; E L represents the total load power; P L_base represents the average power of base load; P L represents the total load average power; max(0, ·) function ensures that the penalty is only applied when the constraint is violated.

[0122] 2) Fitness function reconstruction

[0123] The constraint penalty term is added to the original total cost, and the fitness function is reconstructed by adjusting the penalty coefficient to comprehensively consider cost optimization and constraint satisfaction, and is expressed as: In the formula, f(X) is the fitness value of the honeypot optimization algorithm; C total is the original total cost; λ p is the penalty coefficient.

[0124] ⑤ Configuration scheme generation and verification

[0125] The conventional configuration verification only focuses on static cost indicators, ignores dynamic performance such as renewable energy consumption rate and energy storage efficiency, and lacks robustness test for load fluctuation scenarios, resulting in deviation between actual operation and design expectation.

[0126] After outputting the optimal configuration scheme, the present application performs multi-dimensional verification through typical day scenario simulation. Firstly, the cost deviation rate is calculated to ensure prediction accuracy, then the curtailment rate is tested to meet the renewable energy utilization requirement, and then the energy storage cycle efficiency is verified to meet the performance standard, thereby guaranteeing the robustness of the scheme in actual operation, and the specific process is as follows:

[0127] 1) Optimal configuration scheme output

[0128] The global optimal configuration scheme vector after convergence of the honeypot optimization algorithm is output as the final configuration result, and is expressed as: In the formula, X * is the optimal configuration scheme vector; X represents the configuration vector that makes the fitness function f(X) take the minimum value.

[0129] 2) Cost deviation rate verification

[0130] The relative deviation between the total cost predicted by the optimization algorithm and the running cost based on the typical day scenario simulation is calculated to ensure that the deviation is lower than the preset threshold to verify the cost accuracy, and is expressed as: In the formula, ΔC is the cost deviation rate; C sim (X * ) is the simulation running cost based on the typical day scenario; C pred (X * ) is the total cost predicted by the optimization algorithm; and 5% is the maximum allowable deviation threshold.

[0131] 3) Curtailed power rate verification

[0132] The ratio of the total sum of renewable energy curtailed power to the total sum of actual output is calculated to ensure that the curtailed power rate does not exceed the preset threshold to verify the renewable energy utilization efficiency, and is expressed as: In the formula, is the maximum allowed curtailment rate threshold; P is the renewable energy curtailment power; ∑P is the sum of renewable energy curtailment power in all time periods within the verification period; and P is the renewable energy actual output. re,curt is the maximum allowed curtailment rate threshold; P is the renewable energy curtailment power; ∑P is the sum of renewable energy curtailment power in all time periods within the verification period; and P is the renewable energy actual output. re is the maximum allowed curtailment rate threshold; P is the renewable energy curtailment power; ∑P is the sum of renewable energy curtailment power in all time periods within the verification period; and P is the renewable energy actual output.

[0133] 4) Verification of energy storage cycle efficiency

[0134] The ratio of the total energy discharged by the energy storage to the total energy charged is calculated as the cycle efficiency, and the efficiency is ensured to be higher than the minimum required threshold to verify the performance of the energy storage, which is represented as: In the formula, η ess is the energy storage cycle efficiency; is the energy storage discharge power; is the energy storage charging power; Δt is the time step; and 85% is the minimum efficiency requirement threshold.

[0135] As a specific embodiment that can be implemented, the comprehensive performance of the improved honeypot algorithm of the application in the optimal configuration of an industrial power grid is evaluated, and four groups of comparisons are used in the experiment: a genetic algorithm, a particle swarm algorithm, a traditional honeypot algorithm, and the method of the application.

[0136] The comparison chart of the convergence process of the optimization algorithm is as shown in Figure 3 The method of the application (star-shaped marker line) not only has a better initial solution, but also has a significantly faster convergence speed than other algorithms, and the lowest total cost is obtained under the same number of iterations.

[0137] The comparison chart of the stability of the optimization result is as shown in Figure 4 The distribution of the optimization results of 50 independent runs is shown, the result distribution interval of the method of the application is the narrowest and the mean is the lowest, proving that it is less affected by the initial solution and has strong stability.

[0138] The comparison chart of the calculation efficiency of the optimization algorithm is as shown in Figure 5 The calculation time of the optimization is compared, the method of the application has the shortest time consumption, and the calculation efficiency is obviously improved, which benefits from the adaptive search mechanism driven by the load characteristics.

[0139] The comparison chart of the renewable energy utilization rate of the optimization scheme is as shown in Figure 6 It is shown that the renewable energy consumption rate of the scheme of the application is the highest, which is more than 10 percentage points higher than that of the traditional method, which is due to the adaptive penalty mechanism associated with the load characteristics in the algorithm.

[0140] The comprehensive experiments of the embodiment show that the application realizes the overall improvement of the optimization speed, result quality and stability through the triple improvement of the initialization based on the load mode, the neighborhood search weighted by the fluctuation and the directional disturbance weighted by the similarity.

[0141] As another implementable embodiment, the source-load-storage configuration economic analysis is carried out, and the economy of different configuration schemes is compared from the perspective of the whole life cycle, and the experimental data show the initial investment cost, annual operation cost and annual maintenance cost of four schemes, and the total cost of 10 years is calculated (red dotted line). The scheme of the application has slightly higher initial investment, but with significantly reduced annual operation cost, the total cost of 10 years is the lowest, as shown in the following figure. Figure 7

[0142] The experimental results verify the effectiveness of the double-layer optimization framework of the application. The upper-layer configuration optimization fully considers the coupling effect of the lower-layer operation strategy, and realizes the balance between the whole life cycle cost optimization and environmental protection performance through the source-load-storage collaborative design.

[0143] S5: Grid energy storage collaborative optimization configuration: based on the optimization results, the output is the energy storage capacity optimization configuration scheme considering the lowest whole life cycle cost and efficient renewable energy consumption.

[0144] In the application, based on the optimization results, the actual engineering configuration of the power grid is realized, and the specific implementation process is as follows:

[0145] 1) Configuration scheme deployment: convert the generated optimal configuration scheme vector into entity equipment selection;

[0146] For a photovoltaic system: determine the component quantity and inverter capacity, and complete the array installation combined with the factory roof / empty land layout;

[0147] For a wind turbine system: select the wind turbine model and tower height according to , and arrange the machine in the wind resource rich area of the factory area;

[0148] For an energy storage system: configure lithium batteries or flow batteries according to (capacity) and (power), and connect a bidirectional converter in parallel.

[0149] 2) Collaborative operation strategy embedding: write the optimal operation strategy S * of the lower-layer optimization of S3 into the power grid energy management system (EMS), including:

[0150] Source-load matching mechanism: real-time call the base load / adjustable / impact load mode characteristics extracted by S2, dynamically adjust the photovoltaic / wind turbine output and energy storage charging / discharging plan;

[0151] Multi-objective scheduling: execute the online optimization algorithm of the minimum grid purchase cost, penalty avoidance of abandoned load and suppression of energy storage loss;

[0152] ​​Dynamic feedback adjustment: continuously collect actual energy storage resource data (method same as S1) in the running stage, when it is detected that the load mode deviates significantly from the historical characteristics (by similarity sim k determination), trigger the configuration parameter re-optimization process, ensure that the system always adapts to the dynamic characteristics of the load.

[0153] Finally, a power grid system with "accurate matching of source-load characteristics and dynamic coordination of storage-grid strategy" is formed, and the optimal life cycle cost and efficient consumption of renewable energy are realized.

[0154] The application is a kind of energy storage capacity optimization configuration method based on flexible resource prediction, in use, the application decomposes industrial complex load into typical operation mode with clear physical meaning, improves the accuracy of equipment level energy consumption diagnosis and source load adaptation; the dynamic matching degree evaluation mechanism of the application can reflect the nonlinear coupling characteristics in the actual operation process, ensure that the source load storage system configuration is scientific and adapts to the load fluctuation change; the application improves the comprehensive benefits of the system in multiple dimensions such as investment income, carbon emission reduction contribution and power fluctuation control through multi-objective collaborative optimization; the application uses operation data feedback and evolutionary optimization strategy to adaptively adjust the configuration parameters, and realizes the maintenance of the long-term optimal operation state of the source load storage system; the application has the advantages of constructing a double-layer coupled framework, using an improved honeypot algorithm, constructing an adaptive penalty function, realizing multi-objective dynamic balance, and establishing a closed-loop verification system.

[0155] Example 2

[0156] As Figure 8 shown, a flexible resource prediction-based energy storage capacity optimization configuration system applying the method described above, characterized in that: the system comprises:

[0157] Energy storage resource data acquisition unit: for various production equipment, auxiliary facilities and public loads in industrial plant, real-time data acquisition is carried out through intelligent electric meters deployed at key nodes of power distribution network and plant energy management system;

[0158] Multi-modal feature decomposition and clustering unit: for accurately stripping the base load, adjustable load and impact load of the load by improved variational modal decomposition and Pearson correlation screening;

[0159] Cooperative optimization model construction unit: for constructing a "upper layer capacity-lower layer operation" double-layer coupled framework, and realizing the coupling optimization of source-load-storage parameters through two-layer cooperative iteration;

[0160] Optimization unit: for realizing fast global optimization based on improved honeypot algorithm driven by load mode category, and cooperating with adaptive penalty mechanism and closed-loop verification system to ensure operation feasibility in planning stage;

[0161] The synergistic optimization configuration output unit is configured to output an energy storage capacity optimization configuration scheme that takes into account the lowest life cycle cost and efficient renewable energy consumption based on the foregoing optimization results.

Claims

1. A method for optimizing energy storage capacity allocation based on flexible resource prediction, characterized in that: The method includes the following steps: S1: Energy storage resource data acquisition based on flexible resources: Real-time data acquisition is carried out for various production equipment, auxiliary facilities and public loads in industrial plants through smart meters deployed at key nodes of the power distribution network and the plant energy management system. S2: Multimodal feature decomposition and clustering of energy storage resource data: By improving variational mode decomposition and Pearson correlation screening, the three modes of load, namely base load, adjustable load, and impact load, are accurately separated; S3: Construction of Source-Load-Storage Coordinated Optimization Model: Construct a two-layer coupled framework of "upper-layer capacity-lower-layer operation" and achieve coupled optimization of source-load-storage parameters through two-layer coordinated iteration; S4: Improved honeypot algorithm for optimization: The improved honeypot algorithm based on load pattern category-driven algorithm achieves fast global optimization, and is combined with an adaptive penalty mechanism and closed-loop verification system to ensure operational feasibility in the planning stage; S5: Grid-storage collaborative optimization configuration: Based on the aforementioned optimization results, output an energy storage capacity optimization configuration scheme that balances the lowest life-cycle cost and efficient renewable energy consumption.

2. The energy storage capacity optimization configuration method based on flexible resource prediction as described in claim 1, characterized in that: The data acquisition of energy storage resources based on flexible resources in S1 includes the following specific acquisition process: Data collection coverage: encompasses all power-consuming units within the factory area, including main equipment of continuously operating production lines, auxiliary equipment that starts and stops intermittently, and sudden impact loads; Data acquisition parameter settings: Collect active power time-series data continuously at a 1-minute sampling interval for at least one complete production cycle, and simultaneously record timestamps, equipment operating status and operating condition labels; Data quality assurance: Outliers are eliminated through cross-validation of redundant metering devices, and the data is transmitted to the central database in real time using the plant's communication network to form a standardized raw load power dataset, providing high-precision input for subsequent multimodal feature decomposition.

3. The method for optimizing energy storage capacity allocation based on flexible resource prediction as described in claim 1, characterized in that: The multimodal feature decomposition and clustering of energy storage resource data in S2 specifically includes the following steps: S2.1: Variational Mode Decomposition Model Construction: Construct an optimization model to ensure that the sum of all mode components equals the original energy storage resource data by minimizing the time derivative constraints of the mode components, thereby decomposing the sub-load mode components; S2.2: Modal component screening: Calculate the Pearson correlation coefficient between each modal component and the original energy storage resource data, retain the components with correlation coefficients exceeding the preset threshold, and ensure that the screened modal components are highly correlated with the original load; S2.3: Typical load mode extraction: The screened modal components are classified into base load mode components, adjustable mode components, and impact mode components, thereby identifying and extracting the characteristics of three typical load modes.

4. The method for optimizing energy storage capacity allocation based on flexible resource prediction as described in claim 1, characterized in that: The construction of the source-load-energy storage collaborative optimization model in S3 specifically includes the following steps: S3.1: Upper-level equipment configuration optimization: Optimize the equipment configuration decision vector, including photovoltaic installed capacity, wind turbine installed capacity, energy storage system capacity and rated power, in order to minimize the sum of equipment investment cost and expected operating cost and determine the optimal equipment configuration parameters; S3.2: Lower-level operation strategy optimization: Given the equipment configuration, optimize the operation strategy vector and determine the optimal scheduling scheme by minimizing the total operating cost of grid power purchase, load abandonment penalty and energy storage cycle loss penalty.

5. The method for optimizing energy storage capacity allocation based on flexible resource prediction as described in claim 1, characterized in that: The improved honeypot algorithm in S4 is optimized by the following steps: S4.1: Improved honeypot algorithm for scout bee initialization: Based on the load mode category, the initial honeypot location is generated by combining the historical mean and standard deviation of photovoltaic and wind turbine output, energy storage ratio coefficient, maximum daily power consumption and maximum power of the load under that mode with random perturbation. S4.2: Adaptive neighborhood search during the follower bee stage: An adaptive mechanism driven by load fluctuation is adopted. First, the load fluctuation weight of each time period is calculated. Then, the adaptive step size factor is generated by combining the honeypot fitness. Finally, the honeypot position is updated by weighting the fluctuation weight. S4.3: Observe the dynamic perturbation mechanism of the bee stage: adopt a directional perturbation mechanism, first calculate the similarity between the current load and the historical pattern, then generate a random perturbation intensity, and apply a similarity-weighted directional perturbation to the honey pot location, and combine global optimal information to achieve local optimal escape; S4.4: Calculation of penalty function for multi-objective constraints: Construct an adaptive penalty mechanism based on load characteristics, dynamically adjust the renewable energy consumption penalty weight based on the proportion of total adjustable load electricity, and adjust the carbon emission penalty weight based on the average power proportion of base load. Then, add the penalty term to the total cost to reconstruct the fitness function to achieve multi-objective collaborative optimization. S4.5: Configuration scheme generation and verification: Multi-dimensional verification is carried out through simulation of typical daily scenarios. First, the cost deviation rate is calculated to ensure the accuracy of the prediction. Then, the curtailment rate is verified to meet the requirements of renewable energy utilization. Finally, the energy storage cycle efficiency is verified to meet the performance standards to ensure the robustness of the scheme in actual operation.

6. The energy storage capacity optimization configuration method based on flexible resource prediction as described in claim 5, characterized in that: The adaptive neighborhood search in the follower bee stage of S4.2 is as follows: S4.21: Volatility weight calculation: Calculate the load volatility weight for each period, and quantify the volatility intensity of different periods based on the ratio of the load standard deviation of that period to the maximum standard deviation of all periods; S4.22: Adaptive step size factor calculation: Based on the ratio of honeypot fitness value to the current maximum fitness value and the baseline step size coefficient, calculate the adaptive step size factor for each dimension; S4.23: Honeypot Location Update: Update the honeypot location by adding an adaptive step size factor to the current value and multiplying it by the difference from the random honeypot, and then combining it with the fluctuation weight to achieve a neighborhood search operation weighted by load fluctuation.

7. The method for optimizing energy storage capacity allocation based on flexible resource prediction as described in claim 6, characterized in that: The dynamic perturbation mechanism in the observation bee phase of S4.3 is as follows: S4.31: Load pattern similarity calculation: Calculate the Euclidean distance between the current load curve and each historical pattern curve, and convert it into a similarity value; S4.32: Disturbance intensity generation: Randomly generate disturbance intensity coefficients and distribute them evenly within a preset range to control the disturbance magnitude and enhance the flexibility of the algorithm; S4.33: Honeypot position perturbation: Apply perturbation to the position of the honeypot. The perturbation amount is the perturbation intensity multiplied by the difference between the global optimal honeypot and the current honeypot. Combined with similarity, a similarity-weighted directional perturbation is achieved to escape the local optimum.

8. The method for optimizing energy storage capacity allocation based on flexible resource prediction as described in claim 7, characterized in that: The penalty function calculation for the multi-objective constraints in S4.4 is as follows: S4.41: Calculation of constraint penalty terms: Calculate constraint penalty terms based on the weighted proportion of total adjustable load power based on the degree of insufficient renewable energy consumption and the weighted proportion of average base load power based on the degree of carbon emission exceedance, and dynamically adjust the penalty weights to handle multi-objective constraints; S4.42: Fitness Function Reconstruction: The constraint penalty term is added to the original total cost, and the fitness function is reconstructed by adjusting the penalty coefficient to comprehensively consider cost optimization and constraint satisfaction.

9. The energy storage capacity optimization configuration method based on flexible resource prediction as described in claim 8, characterized in that: The configuration scheme generation and verification in S4.5 are as follows: S4.51: Optimal Configuration Scheme Output: Output the vector of the globally optimal configuration scheme after the honeypot optimization algorithm converges, as the final configuration result; S4.52: Cost Deviation Rate Verification: Calculate the relative deviation between the total cost predicted by the optimization algorithm and the operating cost based on the simulation of typical daily scenarios, and ensure that the deviation is lower than a preset threshold to verify the accuracy of the cost. S4.53: Curtailment Rate Verification: Calculate the ratio of the total curtailed renewable energy power to the total actual output power, and ensure that the curtailment rate does not exceed the preset threshold to verify the utilization efficiency of renewable energy. S4.54: Energy storage cycle efficiency verification: Calculate the ratio of total energy discharged to total energy charged as the cycle efficiency, and ensure that the efficiency is higher than the minimum required threshold to verify the energy storage performance; The grid-storage collaborative optimization configuration in S5 is as follows: S5.1: Configuration Scheme Deployment: Transform the generated optimal configuration scheme vector into physical device selection; S5.2: Cooperative operation strategy embedding: Write the optimal operation strategy optimized at the lower level into the power grid energy management system.

10. A storage capacity optimization configuration system based on flexible resource prediction, applying the method as described in any one of claims 1-9, characterized in that: The system includes: Energy storage resource data acquisition unit: used to collect real-time data from various production equipment, auxiliary facilities and public loads in industrial plants through smart meters deployed at key nodes of the power distribution network and the plant energy management system; Multimodal feature decomposition and clustering unit: used to accurately separate the three modes of load—base load, adjustable load, and impact load—by improving variational mode decomposition and Pearson correlation; Collaborative optimization model building unit: used to build a two-layer coupled framework of "upper-layer capacity-lower-layer operation", and realize the coupled optimization of source-load-storage parameters through two-layer collaborative iteration; Optimization Unit: Used to achieve fast global optimization of the improved honeypot algorithm based on load pattern category, and combined with an adaptive penalty mechanism and closed-loop verification system to ensure operational feasibility during the planning stage; Collaborative optimization configuration output unit: Based on the aforementioned optimization results, it outputs an energy storage capacity optimization configuration scheme that balances the lowest life-cycle cost and efficient renewable energy consumption.

Images