A wind storage capacity and dispatch robust joint optimization method and system

Abstract

The embodiment of the application discloses a wind storage capacity and scheduling robust joint optimization method, adopts a hybrid real number coding particle to simultaneously represent storage capacity configuration and scheduling strategy control parameters; generates a multi-scenario wind / price uncertainty scenario set based on Latin hypercube sampling, generates a complete annual random scenario through cross-period random pairing; and takes the probability weighted expectation of the LP scheduling simulation result in all scenarios as the fitness evaluation basis. Taking multi-scenario expected income, conditional risk value and battery attenuation cost as three optimization objectives, a fast non-dominated sorting and crowding distance mechanism is used to construct a Pareto front; a mutation strategy triggered by an adaptive inertia weight and a crowding distance is used for particle updating; finally, the TOPSIS method is used to recommend an optimal compromise scheme from the Pareto front, solving the problems of low search efficiency of the enumeration method, dependence on single-year data, lack of consideration of uncertainty and single economic index unable to meet multi-dimensional decision-making requirements.

Description

Technical Field

[0001] This invention belongs to the technical field of wind farm energy storage system planning and operation optimization technology, and relates to a robust joint optimization method and system for wind storage capacity and scheduling. Background Technology

[0002] Capacity configuration and operation scheduling of wind-storage combined power generation systems are two deeply coupled decision problems. Capacity configuration (rated power and capacity of energy storage) determines the physical potential and investment cost of the system, while the operation scheduling strategy determines whether this potential can be utilized efficiently and economically. Therefore, the joint optimization of the two is a typical bilevel programming problem: the upper level solves the capacity configuration, and the lower level optimizes the operation strategy under a given configuration.

[0003] Existing representative techniques employ a fixed-step traversal of candidate capacity combinations and rely on deterministic historical data throughout the year for scheduling simulation and economic evaluation. While this method has shown some effectiveness in engineering practice, it suffers from the following fundamental technical shortcomings: (1) The inherent contradiction between search efficiency and resolution. The search cost of the enumeration method is closely related to the step size. High-precision search will cause the number of candidate solutions to increase quadratically, and the computation time will increase from several minutes to tens of minutes. More importantly, a fixed step size grid may completely miss the true optimal solution located between grid points, resulting in an inherent resolution bias.

[0004] (2) Lack of systematic consideration of forecast uncertainty. Existing methods rely on actual data from a single historical year, and their optimization results are "post-hoc" best fits for that specific year. However, actual wind conditions and electricity prices have significant uncertainties, and the "optimal" solution obtained based on single-year data may have severely deteriorated economic performance in years that differ greatly from historical patterns. Although averaging data from multiple years can mitigate this, it will multiply the computational burden and lack a systematic and robust design framework.

[0005] (3) A single economic objective cannot reflect the needs of multidimensional decision-making. Existing methods use the internal rate of return (IRR) as the sole evaluation indicator, but actual engineering decisions usually involve trade-offs between multiple objectives. For example, larger capacity energy storage may bring a higher IRR, but it also means higher initial investment and a wider range of return fluctuations (risks); at the same time, long-term indicators such as battery degradation rate and system utilization rate are also complexly related to economics. When multiple candidate solutions have similar IRRs but different risk characteristics, decision-makers lack transparent multi-criteria comparison tools.

[0006] To this end, this invention proposes a robust joint optimization method for wind storage capacity and scheduling based on an adaptive multi-objective particle swarm optimization algorithm, in order to achieve more efficient, robust and comprehensive joint optimization of wind storage capacity and scheduling. Summary of the Invention

[0007] The purpose of this invention is to solve the problems in the prior art and provide a robust joint optimization method and system for wind storage capacity and scheduling.

[0008] To achieve the above objectives, the present invention employs the following technical solution: A robust joint optimization method for wind storage capacity and scheduling includes the following steps: Obtain historical prediction error distributions of wind farm output forecast data and electricity price forecast data, and generate multiple annual time series scenarios based on the historical prediction error distributions to characterize the uncertainty of wind power output and electricity prices; Construct hybrid coded particles, define the energy storage capacity configuration parameters and scheduling strategy control parameters as the encoding vectors of the hybrid coded particles, and generate an initial population containing multiple particles based on the hybrid coded particles; Based on the initial population, multiple time-series scenarios throughout the year are traversed to obtain the operating benefits under each scenario. Based on the operating benefits, multiple optimization target values ​​of particles corresponding to each scenario are calculated. The multiple optimization target values ​​of each particle are sorted hierarchically, and the crowding distance of each particle is calculated. Based on the hierarchical sorting results, update the individual historical optimal solution of each particle, select a global leader for each particle from the external archive based on the crowding distance, and update the velocity and position of the particles. Based on the updated particles, obtain the particle that currently represents the non-dominated solution and store it in the external archive. When the convergence criterion is met, generate the current non-dominated solution set as the optimal solution set. The optimal solution set is weighted and sorted, and the relative proximity of each scheme is calculated. The scheme with the highest relative proximity is selected as the wind storage capacity configuration and scheduling strategy.

[0009] A further improvement of the present invention is that: The historical prediction error distribution of wind farm output prediction data and electricity price prediction data is obtained, and multiple annual time-series scenarios are generated based on the historical prediction error distribution to characterize the uncertainty of wind power output and electricity price, including: Historical prediction error data of wind farms were obtained, and probability density functions of wind power prediction error and electricity price prediction deviation for each time period were constructed using kernel density estimation. Based on the probability density function, the Latin hypercube sampling method is used to divide and sample the error distribution of each time period into equal probability intervals, generating multiple time-series scenes throughout the year.

[0010] The energy storage capacity configuration parameters include rated energy storage power and rated energy storage capacity; The scheduling strategy control parameters include the prediction assessment penalty weight coefficient, risk avoidance coefficient, and terminal charge state target value in the scheduling model.

[0011] The particle's multiple optimization objectives include annual expected incremental revenue, conditional risk value, and annual battery equivalent degradation cost. The expected annual incremental revenue is calculated based on the probability weighting of operational revenue under all scenarios; The conditional value at risk is used to quantify the downside risk to returns under extremely adverse scenarios. The annual equivalent battery degradation cost is calculated based on the probability-weighted average of the marginal degradation cost of energy storage and the annual equivalent cycle power for each scenario.

[0012] After updating the particle's velocity and position, the process also includes: Based on the crowding distance distribution of the Pareto front in the external archives, determine whether local aggregation has occurred at the front. If so, apply Gaussian perturbation to some particles in the population, and determine whether the mutated particles are retained based on the dominance relationship or simulated annealing criterion.

[0013] The convergence determination criteria include: Calculate the generation distance change rate between adjacent generations of non-dominated solution sets. When the generation distance change rate is less than the preset convergence accuracy threshold for multiple consecutive generations or when the current iteration generation reaches the maximum iteration generation, convergence is determined, and the current non-dominated solution set is generated as the optimal solution set.

[0014] A robust joint optimization system for wind storage capacity and scheduling includes: The uncertainty scenario generation module is used to obtain the historical prediction error distribution of wind farm output prediction data and electricity price prediction data, and generate multiple year-round time series scenarios based on the historical prediction error distribution to characterize the uncertainty of wind power output and electricity price. The population initialization module is used to construct hybrid encoded particles, define the energy storage capacity configuration parameters and scheduling strategy control parameters as the encoding vectors of the hybrid encoded particles, and generate an initial population containing multiple particles based on the hybrid encoded particles; The multi-objective fitness evaluation module is used to traverse multiple time-series scenarios throughout the year based on the initial population, obtain the running benefits under each scenario, calculate multiple optimization objective values ​​for particles corresponding to each scenario based on the running benefits, sort the multiple optimization objective values ​​of each particle hierarchically, and calculate the crowding distance of each particle. The particle swarm evolution update module is used to update the individual historical optimal solution of each particle based on the hierarchical sorting results, select a global leader for each particle from the external archive based on the crowding distance, and update the velocity and position of the particles. The external file maintenance and convergence determination module is used to obtain the particle representing the non-dominated solution based on the updated particle, store it in the external file, and generate the current non-dominated solution set as the optimal solution set when the convergence determination criterion is met. The optimal solution recommendation module is used to perform weighted sorting of the optimal solution set, calculate the relative proximity of each solution, and select the solution with the highest relative proximity as the wind storage capacity configuration and scheduling strategy.

[0015] A computer program product includes a computer program that, when executed by a processor, implements any of the methods described above.

[0016] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of any of the methods described above.

[0017] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the methods described herein.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a robust joint optimization method for wind storage capacity and scheduling. It obtains historical prediction error distribution and generates multiple year-round time-series scenarios based on this distribution. During the optimization process, it traverses multiple scenarios to calculate the operational benefits under each scenario and uses this to calculate the optimization target value. This ensures that the optimization result is no longer dependent on the actual data of a specific year, but can simultaneously consider multiple possible wind conditions and electricity price scenarios, effectively reducing the risk of severe economic deterioration due to deviations in wind conditions and electricity prices from historical patterns during actual operation. Furthermore, it constructs a hybrid encoded particle containing energy storage capacity configuration parameters and scheduling strategy control parameters, uniformly encoding capacity configuration and scheduling strategies into particle vectors. Based on these particles, an initial population is generated for iterative search, transforming the exploration of candidate solutions from discrete grid traversal to population evolution search in continuous space, decoupling search accuracy from computational cost. The inherent contradictions between different objectives are addressed, avoiding the potential omission of truly optimal solutions by fixed-step grids while improving optimization efficiency. By hierarchically sorting the multiple optimization objective values ​​of each particle and calculating the crowding distance to maintain the diversity of the solution set, a non-dominated solution set representing the trade-offs between different objectives is generated as the optimal solution set. This preserves multiple Pareto-efficient solutions where each objective is not mutually dominant, and can demonstrate the objective trade-offs between different dimensions such as benefits and risks without pre-setting subjective weights for each objective. At the same time, an external archive is introduced to store the non-dominated solutions generated in each generation and selects a global leader for the particles based on the crowding distance during the update process, guiding the population to evolve towards a uniform distribution in the Pareto front, avoiding excessive aggregation of the population in local areas, and ensuring that the final Pareto optimal solution set has good distribution in the objective space. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an overall flowchart of a robust joint optimization method for wind storage capacity and scheduling disclosed in an embodiment of the present invention; Figure 2 This is a schematic diagram of uncertainty scenario generation based on Latin hypercube sampling disclosed in an embodiment of the present invention. It is used to demonstrate the full-process generation architecture of scenarios facing two types of uncertain factors: wind power and electricity price. The architecture is divided into four levels from top to bottom: data input layer, probability modeling layer, sampling and synthesis layer, and output result module. Figure 3 This is a schematic diagram of a hybrid encoded particle structure, with a total dimension of D=5: Including capacity parameters ( , ) and scheduling control parameters ( , , ). Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0024] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0025] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0026] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0027] First, the relevant abbreviations and key terms involved in this embodiment will be explained: Particle Swarm Optimization (PSO) is a metaheuristic optimization algorithm based on swarm intelligence. In this invention, PSO is used to efficiently search for the key control parameters of the optimal power-capacity configuration scheme and scheduling strategy of a wind-storage system within a continuous solution space, replacing the enumeration-based search in traditional methods where the computational cost increases quadratically with the accuracy requirements.

[0028] Pareto Optimality: A criterion for determining non-dominated solutions in multi-objective optimization. A solution is considered Pareto optimal if no other feasible solution is superior to a particular solution in all objectives and, at least in one objective, strictly outperforms it. This invention utilizes this theory to construct a non-dominated frontier for three objectives: energy storage revenue, risk, and degradation cost. It quantifies the objective trade-offs between these objectives, providing an objective solution set for engineering decisions without pre-selective subjective weighting.

[0029] Latin Hypercube Sampling (LHS) is a hierarchical sampling strategy that divides the probability distribution of a random variable into non-overlapping intervals and ensures that each interval is sampled, thereby achieving more uniform and comprehensive coverage of the entire distribution space with fewer sample points. In this invention, LHS is used to generate multidimensional stochastic scenarios characterizing the uncertainty of wind power output and electricity prices.

[0030] Crowding Distance: A metric used to measure the density of other solutions around a solution on the Pareto front in a multi-objective optimization algorithm. This invention utilizes this metric to maintain the diversity of the external archive, preferentially retaining solutions located in sparse regions of the front to prevent excessive local clustering of the solution set.

[0031] Conditional Value at Risk (CVaR) is a risk metric that represents the average loss of a portfolio or strategy when its potential loss exceeds its value at risk, given a confidence level (e.g., 95%). This invention uses CVaR as an independent optimization objective to quantify and minimize the downside risk of returns under extremely adverse scenarios.

[0032] The Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) is a multi-criteria decision analysis method that ranks candidate solutions by calculating the weighted distances between each candidate solution and the positive ideal solution (optimal across all criteria) and the negative ideal solution (worst across all criteria). This invention applies it to the Pareto front to automatically select recommended solutions that balance multiple objectives.

[0033] The present invention will now be described in further detail with reference to the accompanying drawings: See Figures 1 to 3 This invention discloses a robust joint optimization method for wind power storage capacity and scheduling, specifically involving a joint optimization method that integrates adaptive multi-objective particle swarm optimization, scenario-driven robust evaluation, and multi-criteria decision-making. This method is particularly suitable for the design of offshore wind farm energy storage systems to cope with highly uncertain environments. The objectives of this embodiment include: (1) Improve search efficiency and accuracy. By utilizing PSO's intelligent search capability in continuous space, the number of candidate solutions and search accuracy are decoupled, overcoming the "step size-cost" dilemma of the enumeration method.

[0034] (2) Enhance the robustness of decision-making. Introduce a multi-scenario probabilistic evaluation system based on LHS to ensure that the optimal solution has statistically stable economic performance when facing uncertainties in wind power output and electricity prices.

[0035] (3) Construct a multi-objective transparent decision-making framework. Establish a Pareto optimization framework with three objectives of "return-risk-decay" and supplement it with the TOPSIS method to provide decision-makers with objective and comprehensive trade-off information and final recommended solutions.

[0036] Example 1 This invention discloses a robust joint optimization method for wind storage capacity and scheduling, comprising: Simultaneous characterization of energy storage capacity configuration using hybrid real-number encoded particles ( , ) and scheduling strategy control parameters ( , , ); The method for generating a multi-scenario wind / electricity price uncertainty scenario set based on Latin hypercube sampling is as follows: kernel density estimation is performed on historical prediction errors by month and prediction scale respectively; the error distribution of each time period is divided into equal probability intervals and sampled independently; a complete year-round random scenario is generated by cross-time period random pairing; and the probability-weighted expectation of LP scheduling simulation results under all scenarios is used as the basis for fitness evaluation.

[0037] With multi-scenario expected return, conditional value at risk (CVaR), and battery degradation cost as the three optimization objectives, a Pareto front is constructed using fast non-dominated sorting and congestion distance mechanisms. Particles are updated using a mutation strategy triggered by adaptive inertia weights and crowding distance; finally, the optimal compromise is recommended from the Pareto front using the TOPSIS method.

[0038] Furthermore, it also includes an adaptive diversity protection mechanism triggered by crowding distance, specifically including: The crowding distance distribution of Pareto front solutions in external archives is periodically evaluated; mutation is triggered when the unevenness of distribution exceeds a threshold; simulated annealing criterion is used to control the acceptance probability of inferior solutions after mutation, so as to dynamically balance the convergence and diversity of the population.

[0039] Furthermore, it also includes a boundary reflection repair method using hybrid encoded particles: For particle dimensions that exceed the boundary after the update, use and The reflection strategy is used to repair the particle motion direction information.

[0040] Furthermore, it also includes a dual-threshold convergence criterion based on generational distance GD: Simultaneously monitor the rate of change of GD and the maximum number of iterations over multiple consecutive generations, and terminate the algorithm when either condition is met, thus balancing solution quality and computational efficiency.

[0041] Furthermore, it also includes efficient solution methods for inner-layer LP scheduling simulations: The annual problem is decomposed into monthly sub-problems, and a highly sparse constraint matrix is ​​stored and solved using the CSR format to achieve fast solution of annual scheduling for a single scenario.

[0042] Example 2 This embodiment describes the specific execution process of each step based on embodiment 1: Step 1: Generation of Uncertainty Scenarios Based on Latin Hypercube Sampling Step 1.1: Historical Statistical Modeling of Prediction Error Historical data from wind farms for at least the past 36 months were obtained. Statistical analyses were performed on wind power forecast errors and day-ahead-real-time electricity price deviations, categorized by different months and forecast time scales. For each month and each forecast time scale, a nonparametric probability density function was constructed for each set of error samples using kernel density estimation (KDE). (H typical value is 48, corresponding to 24-hour forecast, time resolution 0.5 hours) and This is to accurately capture the skewness and heavy-tailed characteristics that may exist in its actual distribution.

[0043] Step 1.2: Latin hypercube sampling for random scene generation Set the total number of scenarios S (in this embodiment, to balance statistical representativeness and computational load, the total number of scenarios is set to 200). For each scenario s (s=1,...,S), perform the following operations: Step 1.2.1: For each scheduling period t (there are N periods throughout the year), use the constructed KDE to divide the probability distribution of wind power prediction error and day-ahead-real-time electricity price deviation into S equal probability intervals; Step 1.2.2: Uniformly and randomly sample once within each interval, and randomly pair the sampled values ​​of each time period t across time periods to finally generate a complete annual wind power prediction error sequence for each scenario s. Electricity price deviation series ; Step 1.2.3: Compare these error / bias sequences with known deterministic day-ahead wind power forecasts. And the recent clearing of electricity prices The data are overlaid to generate S groups of complete year-round time-series scenarios for subsequent scheduling simulations. :

[0044]

[0045] Step 2: Hybrid Encoding Particle Structure Design and Population Initialization Step 2.1: Hybrid Encoding Particle Structure Each particle represents a complete set of energy storage configuration and scheduling strategies, using hybrid real-number encoding, with a total dimension of D=5:

[0046] in, Indicates the rated power of energy storage The range of values ​​is , This refers to the rated power of the wind farm. Indicates the rated capacity of energy storage The range of values ​​is ; This represents the weighting coefficient of predicted performance penalties in the scheduling objectives. The value range is [0.5, 2]. This represents the risk aversion coefficient in the scheduling strategy. The value range is [0, 1]; This represents the target value of the energy storage state of charge as desired by the dispatch terminal. The value range is [0.2, 0.5].

[0047] Step 2.2: Population Initialization Population size set to (In this embodiment, the population size is set to 40). The Latin hypercube method is used for initialization to ensure that the initial population is uniformly distributed in the five-dimensional solution space, avoiding aggregation that may be caused by random initialization.

[0048] Step 2.3: Constraint Handling For particles that exceed the boundary during initialization or subsequent updates, a boundary reflection strategy is used for repair. Use temporary values ;like Use temporary values If the value still exceeds the boundary after reflection, it will be truncated to the nearest boundary value.

[0049] Step 3: Multi-objective fitness evaluation based on linear programming (LP) scheduling simulation For each particle in the population, the process involves three steps: random scenario scheduling simulation, multi-objective index calculation, and Pareto hierarchical sorting. Step 3.1: Solving the inner-layer annual LP scheduling model For each scene s of the i-th particle, its encoded capacity parameter ( , ) and control parameters ( , , Using as input, this algorithm solves a year-round linear programming scheduling optimization problem. The LP model includes 8 types of decision variables and 9 types of energy storage operation constraints to simulate the optimal charging and discharging strategy and operational benefits of energy storage in this scenario. To improve solution efficiency, a sparse matrix CSR format is used for storage and computation.

[0050] Furthermore, the eight decision variables include: energy storage charging power Pch[t], energy storage discharging power Pdch[t], energy storage state of charge (SOC)[t], auxiliary variable for predicting positive deviation assessment penalties fppos[t], auxiliary variable for predicting negative deviation assessment penalties fpneg[t], over-generation recovery variable dppos[t], under-generation recovery variable dpneg[t], and terminal SOC relaxation. ; Furthermore, the nine types of energy storage operation constraints include: power balance, dynamic SOC recursion, upper and lower limits of SOC, upper and lower limits of charging and discharging power, mutual exclusion of charging and discharging, and terminal SOC constraints.

[0051] Step 3.2: Parameterized configuration of the objective function of the inner LP scheduling model In this invention, the three scheduling strategy parameters of particle encoding ( , , Without changing the structure of the LP model, different scheduling behavior preferences are guided by adjusting the weight coefficients of corresponding terms in the objective function. The objective function (minimization) of the inner LP scheduling model is as follows:

[0052] The first item is the charging cost, which includes the cost of purchasing electricity at the real-time electricity price and the marginal degradation cost λ of energy storage. The first term represents the real-time electricity price for time period t; the second term represents the discharge revenue (negative cost), including real-time electricity sales revenue and the marginal degradation cost λ of energy storage; the third term represents the penalty for prediction deviation. The first item represents the penalty coefficient, which is the amount of penalty payable for every 1MW exceeding the allowable threshold for one period of time, expressed in yuan / MW / period, with a typical value of 4.17 yuan / MW / period; the fourth item is the day-ahead-real-time deviation revenue recovery. This represents the over-issuance recovery coefficient for time period t. The fifth item represents the recovery coefficient for period t; it is the risk penalty item. The sixth item is the penalty for deviation of terminal SOC. This is the penalty coefficient. Satisfying constraints and 0.

[0053] Step 3.3: Quantitative Calculation of Three-Dimensional Optimization Objectives Based on scene weighting, construct the three-target fitness vector of this particle. . The expected incremental return for the year is expressed as ; The 5th percentile conditional value at risk (CVaR) quantifies the risk of extreme downside losses. The annual battery equivalent degradation cost is expressed as , This refers to the marginal degradation cost of energy storage (typically 50 yuan / MWh). The equivalent annual cycle power of energy storage for scenario s.

[0054] Step 3.4: Fast Non-Dominated Sort and Crowded Distance Calculation Based on Pareto dominance, the entire population is stratified to generate multi-level non-dominated frontiers. , This is the optimal solution set for the current iteration.

[0055] Let A be any solution with the three-objective fitness vector as follows: (A)= Define the conversion target. (A)= All three objectives are considered to be minimized.

[0056] Now, any solution B is added if and only if If at least one of the above three inequalities is strictly true, then A is said to dominate B; if A does not dominate B, and B does not dominate A, then A and B are said to be mutually non-dominant, and both are Pareto efficient solutions. Within each frontal layer, the three-dimensional targets are sorted, and the crowding distance of each particle within the same frontal layer is calculated to characterize the density of solutions around it.

[0057] Step 4: Adaptive Mutation and Multi-Objective Particle Swarm Optimization Update Step 4.1: Individual and Global Optimal Choice For each particle Individual historical best Update using dominance relationship: If the current particle dominates ,but =Current particle; if they do not dominate each other, retain the original particle with a 50% probability. .

[0058] Global Guide Select from external files: Using the reciprocal of the crowding distance as the selection probability, particles tend to move toward the sparse region of the Pareto front, promoting a uniform distribution of the front.

[0059] Step 4.2: Speed ​​and Position Update The update of particle i in the (k+1)th generation follows the standard PSO formula, as follows:

[0060]

[0061] in, and This is the acceleration factor (typical value 2.05); and For each iteration, a uniformly distributed decimal number in the range [0, 1] is generated in real time as an independent random number. The adaptive convex decreasing inertia weight coefficients are as follows:

[0062] in, Let be the maximum inertia weighting coefficient, and let the maximum inertia weighting coefficient be a typical value of 0.9; Let be the minimum inertia weight coefficient, and let the minimum inertia weight coefficient be a typical value of 0.4; Set the maximum number of iterations to 200.

[0063] Step 4.3: Adaptive Diversity Triggers Mutation Every Δk=10 generations, frontier clustering determination is performed, and the ratio of extreme crowding distances for each target dimension in the external archive is calculated. If the ratio exceeds a threshold... The Pareto front is identified as a local cluster, triggering adaptive mutation: Step 4.3.1: Randomly select 20% of the particles in the population and apply Gaussian perturbation, with the standard deviation of the perturbation being 5% of the corresponding search interval; Step 4.3.2: After re-evaluating the fitness of the mutated particle, if the mutated solution dominates the original particle, it is directly retained; if there is no dominance relationship, the inferior solution is accepted using the simulated annealing probability criterion, and the acceptance probability is... The iteration process T decreases linearly. In the early stage, inferior solutions are retained with a high probability to maintain population diversity, while in the later stage, only improved solutions are accepted.

[0064] Step 5: External File Maintenance and Convergence Determination Step 5.1: External File Update The non-dominated solutions generated in each generation are merged with the external archive and sorted for non-domination. If the archive size exceeds a preset upper limit, redundant individuals are deleted sequentially according to the crowding distance from smallest to largest to ensure that the solutions in the archive can represent the complete Pareto front and are spatially uniformly distributed. In this embodiment, the upper limit is 100.

[0065] Step 5.2: Convergence Determination A dual-threshold convergence criterion is employed to calculate the generational distance (GD) between the optimal frontier of each generation and the previous generation. The convergence is achieved when the rate of change of GD for multiple consecutive generations is less than the convergence accuracy threshold. 10 -4 If the number of iterations reaches the maximum allowed number of iterations, the algorithm is considered converged, and the external archive is output as the final Pareto optimal solution set. In this embodiment, the number of consecutive iterations is 30, and the maximum allowed number of iterations is 200.

[0066] Step 6: Pareto Front Analysis and TOPSIS for Decision Support Step 6.1: Visualization and Trade-off Analysis of the Pareto Front In three-dimensional target space By plotting the complete Pareto front surface and calculating the marginal rate of substitution between adjacent optimal solutions, the objective trade-off between benefits, risks, and decay costs can be visually demonstrated.

[0067] Step 6.2: TOPSIS weighted ranking automatically recommends the optimal solution. For engineering scenarios requiring a single optimal configuration, TOPSIS quantitative ranking is adopted: Step 6.2.1: Normalize the three-dimensional target vector to eliminate dimensional differences; Step 6.2.2: Input the decision-maker's custom weight vector The default weights are [0.33, 0.33, 0.33]. Step 6.2.3: Construct the positive ideal solution Negative ideal solution ; Step 6.2.4: Calculate the weighted standardized value of each scheme and its Euclidean distance from the positive and negative ideal solutions. and According to relative closeness Sort the options, and the one with the largest C value is the recommended option.

[0068] Step 6.3: Supplementary calculation of financial indicators for the recommended scheme The optimal particle x recommended by TOPSIS It supplements the calculation of a complete financial indicator system, such as initial investment, operation and maintenance costs, internal rate of return, net present value and static investment payback period, to form a detailed decision report for engineering applications.

[0069] This embodiment discloses a collaborative optimization architecture combining PSO continuous search and LP precise scheduling. It upgrades the outer capacity configuration problem from enumeration to PSO-driven continuous spatial intelligent search, fundamentally resolving the contradiction between search accuracy and computational cost. Simultaneously, the particle encoding incorporates both capacity and scheduling strategy parameters, achieving collaborative optimization of physical configuration and operational strategy.

[0070] Furthermore, a robust fitness evaluation system based on LHS multi-scenario driving is proposed. A representative scenario set is generated using LHS, with expected return, CVaR risk, and battery degradation cost jointly forming the objective function, directly addressing uncertainty in the optimization process. This system can quantify and balance returns and extreme risks, generating solutions that are statistically robust to prediction errors.

[0071] Furthermore, an adaptive diversity maintenance and premature convergence prevention mechanism is implemented. Adaptive mutation is dynamically triggered by monitoring the crowding distance distribution at the Pareto front. This mechanism effectively maintains population diversity and avoids the algorithm getting trapped in local optima without significantly increasing computational burden.

[0072] Furthermore, a boundary reflection repair strategy for hybrid encoded particles is employed. By using reflection instead of truncation to handle out-of-bounds particles, the motion trend information of the particles is preserved, which helps to explore better feasible solutions near the constraint boundary.

[0073] Example 3 This invention discloses a robust joint optimization system for wind storage capacity and scheduling, comprising: The uncertainty scenario generation module is used to obtain the historical prediction error distribution of wind farm output prediction data and electricity price prediction data, and generate multiple year-round time series scenarios based on the historical prediction error distribution to characterize the uncertainty of wind power output and electricity price. The population initialization module is used to construct hybrid encoded particles, define the energy storage capacity configuration parameters and scheduling strategy control parameters as the encoding vectors of the hybrid encoded particles, and generate an initial population containing multiple particles based on the hybrid encoded particles; The multi-objective fitness evaluation module is used to traverse multiple time-series scenarios throughout the year based on the initial population, obtain the running benefits under each scenario, calculate multiple optimization objective values ​​for particles corresponding to each scenario based on the running benefits, sort the multiple optimization objective values ​​of each particle hierarchically, and calculate the crowding distance of each particle. The particle swarm evolution update module is used to update the individual historical optimal solution of each particle based on the hierarchical sorting results, select a global leader for each particle from the external archive based on the crowding distance, and update the velocity and position of the particles. The external file maintenance and convergence determination module is used to obtain the particle representing the non-dominated solution based on the updated particle, store it in the external file, and generate the current non-dominated solution set as the optimal solution set when the convergence determination criterion is met. The optimal solution recommendation module is used to perform weighted sorting of the optimal solution set, calculate the relative proximity of each solution, and select the solution with the highest relative proximity as the wind storage capacity configuration and scheduling strategy.

[0074] A schematic diagram of a terminal device according to an embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0075] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention.

[0076] The terminal device can be a desktop computer, laptop computer, cloud server, or other device with strong computing power. The terminal device may include, but is not limited to, a processor and memory.

[0077] The optimal choice for the processor is a multi-core high-speed central processing unit (CPU).

[0078] The memory can be used to store the computer program and / or module. The processor implements various functions of the terminal device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0079] If the modules / units integrated into the terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0080] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A robust joint optimization method for wind storage capacity and scheduling, characterized in that, Includes the following steps: Obtain historical prediction error distributions of wind farm output forecast data and electricity price forecast data, and generate multiple annual time series scenarios based on the historical prediction error distributions to characterize the uncertainty of wind power output and electricity prices; Construct hybrid coded particles, define the energy storage capacity configuration parameters and scheduling strategy control parameters as the encoding vectors of the hybrid coded particles, and generate an initial population containing multiple particles based on the hybrid coded particles; Based on the initial population, multiple time-series scenarios throughout the year are traversed to obtain the operating benefits under each scenario. Based on the operating benefits, multiple optimization target values ​​of particles corresponding to each scenario are calculated. The multiple optimization target values ​​of each particle are sorted hierarchically, and the crowding distance of each particle is calculated. Based on the hierarchical sorting results, update the individual historical optimal solution of each particle, select a global leader for each particle from the external archive based on the crowding distance, and update the velocity and position of the particles. Based on the updated particles, obtain the particle that currently represents the non-dominated solution and store it in the external archive. When the convergence criterion is met, generate the current non-dominated solution set as the optimal solution set. The optimal solution set is weighted and sorted, and the relative proximity of each scheme is calculated. The scheme with the highest relative proximity is selected as the wind storage capacity configuration and scheduling strategy.

2. The robust joint optimization method for wind storage capacity and scheduling according to claim 1, characterized in that, The historical prediction error distribution of wind farm output prediction data and electricity price prediction data is obtained, and multiple annual time-series scenarios are generated based on the historical prediction error distribution to characterize the uncertainty of wind power output and electricity price, including: Historical prediction error data of wind farms were obtained, and probability density functions of wind power prediction error and electricity price prediction deviation for each time period were constructed using kernel density estimation. Based on the probability density function, the Latin hypercube sampling method is used to divide and sample the error distribution of each time period into equal probability intervals, generating multiple time-series scenes throughout the year.

3. The robust joint optimization method for wind storage capacity and scheduling according to claim 1, characterized in that, The energy storage capacity configuration parameters include rated energy storage power and rated energy storage capacity; The scheduling strategy control parameters include the prediction assessment penalty weight coefficient, risk avoidance coefficient, and terminal charge state target value in the scheduling model.

4. The robust joint optimization method for wind storage capacity and scheduling according to claim 1, characterized in that, The particle's multiple optimization objectives include annual expected incremental revenue, conditional risk value, and annual battery equivalent degradation cost. The expected annual incremental revenue is calculated based on the probability weighting of operational revenue under all scenarios; The conditional value at risk is used to quantify the downside risk to returns under extremely adverse scenarios. The annual equivalent battery degradation cost is calculated based on the probability-weighted average of the marginal degradation cost of energy storage and the annual equivalent cycle power for each scenario.

5. The robust joint optimization method for wind storage capacity and scheduling according to claim 1, characterized in that, After updating the particle's velocity and position, the process also includes: Based on the crowding distance distribution of the Pareto front in the external archives, determine whether local aggregation has occurred at the front. If so, apply Gaussian perturbation to some particles in the population, and determine whether the mutated particles are retained based on the dominance relationship or simulated annealing criterion.

6. The robust joint optimization method for wind storage capacity and scheduling according to claim 1, characterized in that, The convergence determination criteria include: Calculate the generation distance change rate between adjacent generations of non-dominated solution sets. When the generation distance change rate is less than the preset convergence accuracy threshold for multiple consecutive generations or when the current iteration generation reaches the maximum iteration generation, convergence is determined, and the current non-dominated solution set is generated as the optimal solution set.

7. A robust joint optimization system for wind storage capacity and scheduling, characterized in that, include: The uncertainty scenario generation module is used to obtain the historical prediction error distribution of wind farm output prediction data and electricity price prediction data, and generate multiple year-round time series scenarios based on the historical prediction error distribution to characterize the uncertainty of wind power output and electricity price. The population initialization module is used to construct hybrid encoded particles, define the energy storage capacity configuration parameters and scheduling strategy control parameters as the encoding vectors of the hybrid encoded particles, and generate an initial population containing multiple particles based on the hybrid encoded particles; The multi-objective fitness evaluation module is used to traverse multiple time-series scenarios throughout the year based on the initial population, obtain the running benefits under each scenario, calculate multiple optimization objective values ​​for particles corresponding to each scenario based on the running benefits, sort the multiple optimization objective values ​​of each particle hierarchically, and calculate the crowding distance of each particle. The particle swarm evolution update module is used to update the individual historical optimal solution of each particle based on the hierarchical sorting results, select a global leader for each particle from the external archive based on the crowding distance, and update the velocity and position of the particles. The external file maintenance and convergence determination module is used to obtain the particle representing the non-dominated solution based on the updated particle, store it in the external file, and generate the current non-dominated solution set as the optimal solution set when the convergence determination criterion is met. The optimal solution recommendation module is used to perform weighted sorting of the optimal solution set, calculate the relative proximity of each solution, and select the solution with the highest relative proximity as the wind storage capacity configuration and scheduling strategy.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-6.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

Images