An industrial park optimal scheduling method based on wild horse swarm algorithm

Abstract

This invention discloses an industrial park optimization scheduling method based on the wild horse herd algorithm, comprising: establishing a multi-energy resource model of the park and an objective function oriented towards minimizing the overall expected operating cost of the industrial park; constructing a multivariate joint probability distribution model of wind power, photovoltaic power, and load using the R-Teng Copula method to quantify source-load uncertainty and generate multiple typical scenarios; establishing a two-layer optimization model, with the upper layer taking the minimization of the overall expected operating cost of the industrial park as the global optimization objective, and the lower layer dividing the park into several aggregation units with the minimization of the operating cost of each unit as the local optimization objective; performing sub-problem optimization for each aggregation unit based on the wild horse herd algorithm and ADMM; and achieving optimized scheduling based on the charging and discharging power of energy storage units, the demand response reduction power of flexible units, and the output value of distributed units in each aggregation unit obtained through optimization.

Description

Technical Field

[0001] This invention belongs to the field of integrated energy system optimization scheduling and energy management technology in industrial parks, and relates to an optimization scheduling method for industrial parks based on the wild horse herd algorithm. Background Technology

[0002] Industrial parks, as key scenarios for multi-energy coupling and cascade utilization in integrated energy systems, integrate fluctuating power sources such as distributed photovoltaics and wind power, as well as various types of regulating resources such as energy storage systems and flexible loads. Within these systems, multiple energy media, including electricity, heat, and gas, are deeply intertwined in the production, conversion, and consumption stages, exhibiting complex physical characteristics characterized by high dimensionality, strong coupling, and multiple constraints. This makes the characterization and optimization decisions of the overall system's operating domain extremely difficult.

[0003] In practical operation, industrial park scheduling strategies need to accurately identify extreme operating scenarios, ensuring the renewable energy absorption rate while also considering the system's robustness and economy, thereby providing effective decision support for multi-entity collaborative scheduling. Furthermore, the introduction of non-convex / mixed integer constraints such as mutual exclusion of energy storage charging and discharging and equipment start-up and shutdown makes existing scheduling strategies prone to getting trapped in local optima when searching for the global optimum.

[0004] However, existing technologies generally have the following shortcomings:

[0005] (1) Insufficient characterization of source-load multidimensional uncertainty. The wind and solar power output and load have both non-Gaussian characteristics of their respective marginal distributions and significant temporal and intervariate correlations. If the independence assumption or simple correlation coefficient model is adopted, it is easy to cause scene deviation, and the robustness and feasibility of the scheduling scheme are insufficient.

[0006] (2) Centralized scheduling is difficult to scale. The park's resources are scattered and there are many entities involved, making it difficult to balance real-time performance, privacy and scalability in a centralized large-scale optimization solution.

[0007] (3) Problems involving non-convex and mixed integers are difficult to solve. The mutual exclusion of energy storage charging and discharging, equipment start-up and shutdown, and segmented / discrete demand response make the optimization problem exhibit non-convex or mixed integer characteristics, which limits the applicability of traditional gradient methods or single programming methods.

[0008] (4) Insufficient risk control and robustness. If the prediction error of new energy sources is not constructed with an uncertainty set, the scheduling scheme may violate power balance or equipment constraints, reducing the safety of system operation. Summary of the Invention

[0009] To overcome the problems of insufficient characterization of source-load correlation, difficulty in extending centralized optimization, and difficulty in solving non-convex mixed integer problems in existing technologies, this invention provides an industrial park optimization scheduling method based on the wild horse herd algorithm.

[0010] This invention provides an optimized scheduling method for industrial parks based on the wild horse herd algorithm, comprising:

[0011] Step 1: Establish a multi-energy resource model for the industrial park and an objective function oriented towards minimizing the overall expected operating cost of the industrial park;

[0012] Step 2: The R-Teng Copula method is used to construct a multivariate joint probability distribution model of wind power, photovoltaics and load to quantify source-load uncertainty and generate multiple typical scenarios;

[0013] Step 3: Establish a two-layer optimization scheduling model. The upper layer takes the minimum expected operating cost of the industrial park as the global optimization objective, while the lower layer divides the park into several aggregation units and takes the minimum operating cost of each unit as the local optimization objective. Based on the wild horse herd algorithm and ADMM, sub-problem optimization is performed for each aggregation unit.

[0014] Step 4: Based on the charging and discharging power of the energy storage units in each aggregation unit, the demand response reduction power of the flexible units, and the output value of the distributed units, optimize scheduling is achieved.

[0015] The present invention provides an industrial park optimization scheduling method based on the wild horse herd algorithm, which has the following beneficial effects:

[0016] 1. By establishing a source-load joint distribution through R-vine Copula, the non-Gaussian properties of the edges and the multi-dimensional correlation structure can be characterized simultaneously, improving the realism of typical scenarios and enhancing the robustness of scheduling schemes.

[0017] 2. A two-layer robust-distributed coordination architecture is adopted. The upper layer provides the global feasible boundary / signal, and the lower layer solves in parallel and satisfies coupling constraints through ADMM consistency coordination, thereby improving scalability and feasibility.

[0018] 3. Introducing the wild horse herd algorithm in non-convex / discrete scenarios enhances global search capabilities and solution stability, adapting to complex constraints such as energy storage mutual exclusion, demand response segmentation, and equipment start-up and shutdown. Attached Figure Description

[0019] Figure 1 This is a flowchart of an industrial park optimization scheduling method based on the wild horse herd algorithm of the present invention. Detailed Implementation

[0020] like Figure 1 As shown, this invention provides an optimized scheduling method for industrial parks based on the wild horse herd algorithm, comprising:

[0021] Step 1: Establish a multi-energy resource model for the industrial park and an objective function oriented towards minimizing the overall expected operating cost of the industrial park, specifically as follows:

[0022] Step 1.1: Establish a multi-energy resource set R for the park; multi-energy resources include load type and energy type: load type includes flexible load and rigid load, and energy type includes photovoltaic and wind power.

[0023] Step 1.2: Collect multi-source time-series data and perform data cleaning; multi-source time-series data includes wind power output, photovoltaic power output, load, energy storage status and electricity price.

[0024] Step 1.3: Taking into account the cost of electricity purchase and sale, the equivalent lifespan and operation and maintenance costs of energy storage, demand response compensation, wind and solar curtailment penalties, and carbon emission costs, construct an objective function oriented towards minimizing the total expected cost of park operation, specifically as follows:

[0025] Establish the following objective function:

[0026]

[0027] Where t is the scheduling period, A collection of typical scenarios, As scene weight, To cover the cost of purchasing and selling electricity, To reduce charging and discharging operation and maintenance costs, To compensate for the cost of demand response, As punishment for abandoning wind and light, Cost of carbon emissions.

[0028] The expression for the cost of purchasing and selling electricity is:

[0029]

[0030] in, The electricity purchase price for time period t. For the power of the connection line between the park and the power grid, Let t be the electricity price during time period t.

[0031] The expression for charging and discharging operation and maintenance costs is:

[0032]

[0033] in, The unit charging cost coefficient for energy storage systems. For energy storage charging power, This is the unit discharge cost coefficient for the energy storage system; This refers to the energy storage discharge power.

[0034] The expression for the penalty for abandoning wind and light is:

[0035]

[0036] in, The unit is the penalty coefficient for wind and solar power curtailment. For the t-time period scenario The amount of wind and solar power curtailed.

[0037] The carbon emission cost expression is:

[0038]

[0039]

[0040] in, The unit carbon emission penalty coefficient; For the system in the t-time period scenario Total carbon emissions under Calculated from the electricity purchase emission factor and the fuel emission factor, The emission intensity coefficient for electricity purchase. This refers to the unit output release factor of the distributed generation unit. Power generation capacity of distributed generating units.

[0041] Step 1.4: Establish the physical operation boundary model of the park, covering the system power balance constraints throughout the entire time period, the dynamic evolution constraints of the energy storage system's SOC, and the physical output limits of the equipment. This provides a complete mathematical model foundation for subsequent scenario-based uncertainty analysis and two-level optimization scheduling. Specifically:

[0042] The power balance constraint is expressed as:

[0043]

[0044]

[0045] in, Contribute to new energy forecasting For the predicted power of wind power, The predicted power output of photovoltaic power; This refers to the power of wind and solar power that has been curtailed. This refers to the discharge power of the energy storage system. Power for charging energy storage systems; Power generation capacity of distributed generating units; This value represents the power of the connection line with the main power grid. A positive value indicates that electricity is purchased from the grid, while a negative value indicates that electricity is sold to the grid. Base load power refers to the initial electricity demand of the park during the corresponding time period and scenario; Reduce power to meet demand.

[0046] The dynamic constraints of energy storage SOC are:

[0047]

[0048] in, This represents the state of charge of the energy storage system at the end of time period t+1 under scenario ω, and it represents the remaining energy in the energy storage device. This represents the state of charge of the energy storage system at the beginning of time period t under scenario ω, i.e., the initial energy state; This indicates the charging efficiency of the energy storage system. This indicates the discharge efficiency of the energy storage system; This indicates the unit time step of the scheduling.

[0049] And satisfy:

[0050]

[0051] in: This indicates the minimum permissible state of charge for an energy storage system. This indicates the maximum allowable state of charge of the energy storage system, i.e., the rated capacity of the energy storage. This indicates the maximum allowable charging power of the energy storage system. This indicates the maximum allowable discharge power of the energy storage system.

[0052] Step 2: The R-Teng Copula method is used to construct a multivariate joint probability distribution model of wind power, photovoltaics, and load to quantify source-load uncertainty and generate several typical scenarios, specifically:

[0053] Step 2.1: Perform kernel density estimation on the three random variables—wind power output, photovoltaic power output, and flexible load—to obtain the corresponding marginal distribution functions:

[0054]

[0055] in, Let be the marginal distribution function of the j-th random variable, where j=1 represents wind power, j=2 represents photovoltaic power, and j=3 represents flexible load. The marginal probability density function is obtained through kernel density estimation; For kernel functions; It can be any one of the three random variables: wind power output, photovoltaic power output, or flexible load. Let be the sampled data points at time i for the j-th random variable; n is the number of sampled data points; h is the bandwidth, which controls the width of the kernel function.

[0056] Each random variable is transformed into a uniformly distributed variable through uniform transformation. :

[0057]

[0058] in, To minimize the value and avoid boundary issues.

[0059] Step 2.2: Decompose multidimensional dependencies using the tree structure of R-vine Copula, construct a joint probability distribution through a series of two-dimensional Copula functions, and generate a large number of typical scenarios. The joint probability density function is:

[0060]

[0061] in, , is the Copula density function.

[0062] Step 2.3: Use the K-means clustering algorithm to reduce the scene selection and computational complexity. The objective function is:

[0063]

[0064] in, This represents the i-th original scene vector generated by sampling. It includes all data on wind power, solar power, and flexible loads over a period of time; This is the set of scenes contained in the k-th cluster; This represents the center vector of the k-th cluster, used to represent the typical scene features under this cluster.

[0065] A set of typical scenarios was obtained through iterative optimization. This is used for subsequent optimization and scheduling, calculating the weight of each typical scenario. The calculation formula is as follows:

[0066]

[0067] in, This represents the number of original scenes assigned to the k-th cluster; N represents the total number of original scenes generated through sampling. Typical scenarios The weight.

[0068] Step 3: Establish a two-layer optimization scheduling model. The upper layer takes the minimum expected operating cost of the industrial park as the global optimization objective, while the lower layer divides the park into several aggregate units and takes the minimum operating cost of each unit as the local optimization objective. Based on the wild horse herd algorithm and ADMM, sub-problem optimization is performed for each aggregate unit. Due to the existence of mutual exclusion constraints for energy storage charging and discharging and discrete constraints for equipment start-up and shutdown, the sub-problems are non-convex and discontinuous, and traditional gradient methods or second-order cone programming are difficult to solve directly. Therefore, WHO is introduced to improve the success rate of solution and global optimization capability.

[0069] Step 3 specifically involves:

[0070] Step 3.1: Take the typical scenario set generated by the R-vine Copula and the corresponding weights as uncertain inputs, and establish an upper-level robust optimization model with the minimum overall operating expected cost of the industrial park in Step 1 as the optimization objective.

[0071] Step 3.2: Divide the park into several aggregation units. With the minimum operating cost of each unit as the local optimization objective, introduce consistency variables and use ADMM for parallel distributed coordination iteration of each unit to obtain the current decision variables, including the charge-discharge power of the energy storage unit, the demand response reduction power of the flexible unit, and the output value of the distributed generator set. Specifically:

[0072] Divide the park into aggregation units and initialize a set of decision variables for each unit , and introduce consistency variables , and use ADMM for iteration:

[0073]

[0074]

[0075]

[0076] Among them, is the local objective function, corresponding to the operating cost of the i-th dispatchable unit; is the penalty factor; is the coefficient matrix used to map to the global consistency space; is the global consistency variable, k is the number of iterations, is the penalty factor, is the dual variable, reflecting the deviation compensation between the local scheme and the global coordination scheme; obtain the current decision variables of each aggregation unit through the above formula.

[0077] Step 3.3: Judge whether the currently output decision variables meet the convergence condition. If they meet the convergence condition, then the current charge-discharge power of the energy storage unit, the demand response reduction power of the flexible unit, and the output value of the distributed generator set are output as the optimal solutions.

[0078] Specifically in implementation, judge whether to converge according to the following formula:

[0079]

[0080]

[0081]

[0082]

[0083] in, For the original residual, For dual residuals, Original residual tolerance, Dual residual tolerance.

[0084] Step 3.4: If the convergence condition is not met, initiate the wild horse herd algorithm, mapping the decision variables to the positions of individual wild horses. Initialize the wild horse herd for each aggregation unit and calculate the fitness value of each individual wild horse. Use the wild horse herd algorithm WHO as the underlying solver for the ADMM subproblem. That is, in each iteration of ADMM, to solve the subproblem with non-convex constraints, call the wild horse herd algorithm WHO to find the current optimal position. Specifically:

[0085] Step 3.4.1: Initialize the wild horse population. Let the decision vector of the i-th aggregation unit be... The group size is M, and the position of the nth wild horse is... The fitness function is:

[0086]

[0087] in, Let be the fitness value of the nth wild horse individual in the i-th aggregation unit. The smaller the fitness value, the better the scheduling scheme. Let be the local objective function, corresponding to the operating cost of the i-th aggregation unit; To constrain violations and penalties;

[0088] Step 3.4.2: The wild horse herd algorithm iteratively updates the individual positions until the maximum number of iterations or the fitness convergence threshold is reached, obtaining the optimal wild horse individual corresponding to each aggregation unit, i.e., the optimal decision variable.

[0089] Step 3.5: Return the best wild horse individual corresponding to each aggregation unit, that is, the wild horse individual that minimizes the fitness value of each aggregation unit, to step 3.2 so that each aggregation unit can obtain the next generation of decision variables through local iteration using the ADMM algorithm, and determine whether the convergence condition is met. If it is met, output the next generation of decision variables; otherwise, continue to execute steps 3.4 and 3.5 until the convergence condition is met.

[0090] Step 4: Based on the charging and discharging power of the energy storage units in each aggregation unit, the demand response reduction power of the flexible units, and the output value of the distributed units, optimize scheduling is achieved.

[0091] The above description is only a preferred embodiment of the present invention and is not intended to limit the ideas of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An optimized scheduling method for industrial parks based on the wild horse herd algorithm, characterized in that, including: Step 1: Establish a multi-energy resource model for the park and an objective function oriented to minimizing the overall operating expected cost of the industrial park; Step 2: Adopt the R-vine Copula method to construct a multivariate joint probability distribution model of wind power, photovoltaic power and load to quantify the uncertainty of power sources and loads, and generate multiple typical scenarios; Step 3: Establish a two-layer optimal scheduling model. The upper layer takes the minimum overall operating expected cost of the industrial park as the global optimization goal, and the lower layer divides the park into several aggregation units with the minimum operating cost of each unit as the local optimization goal. Based on the wild horse herd algorithm and ADMM, sub- problem optimization is carried out for each aggregation unit; Step 4: According to the charging and discharging power of the energy storage unit, the demand response reduction power of the flexible unit and the output value of the distributed unit obtained by optimization in each aggregation unit, realize the optimal scheduling.

2. The industrial park optimization scheduling method based on the wild horse herd algorithm according to claim 1, characterized in that, The specific content of Step 1 is as follows: Step 1.1: Establish a multi-energy resource set R for the park; the multi-energy resources include load types and energy types: the load types include flexible loads and rigid loads, and the energy types include photovoltaic power and wind power; Step 1.2: Collect multi-source time series data and perform data cleaning; the multi-source time series data includes wind power output, photovoltaic power output, load, energy storage state and electricity price; Step 1.3: Considering the purchase and sale electricity cost, the equivalent life and operation and maintenance cost of the energy storage, the demand response compensation, the penalty for wind and light abandonment and the carbon emission cost comprehensively, construct an objective function oriented to minimizing the total expected cost of park operation; Step 1.4: Establish a physical operation boundary model for the park, covering the system power balance constraint at all times, the dynamic evolution constraint of the energy storage system SOC and the physical output limit of the equipment, providing a complete mathematical model basis for subsequent scenario-based uncertainty analysis and two-layer optimal scheduling.

3. The industrial park optimization scheduling method based on the wild horse herd algorithm according to claim 2, characterized in that, The specific content of Step 1.3 is as follows: Establish the following objective function: Where t is the scheduling period, A collection of typical scenarios, As scene weight, To cover the cost of purchasing and selling electricity, To reduce charging and discharging operation and maintenance costs, To compensate for the cost of demand response, As punishment for abandoning wind and light, For carbon emission costs; The expression of the purchase and sale electricity cost is: in, The electricity purchase price for time period t. For the power of the connection line between the park and the power grid, The electricity price during time period t; The expression of the charging and discharging operation and maintenance cost is: in, The unit charging cost coefficient for energy storage systems. For energy storage charging power, This is the unit discharge cost coefficient for the energy storage system; This refers to the energy storage discharge power; The expression of the penalty for wind and light abandonment is: in, The unit is the penalty coefficient for wind and solar power curtailment. For the t-time period scenario The amount of wind and solar power curtailed; The expression of the carbon emission cost is: in, The unit carbon emission penalty coefficient; For the system in the t-time period scenario Total carbon emissions under Calculated from the electricity purchase emission factor and the fuel emission factor, The emission intensity coefficient for electricity purchase. This refers to the unit output release factor of the distributed generation unit. Power generation capacity of distributed generating units.

4. The industrial park optimization scheduling method based on the wild horse herd algorithm according to claim 3, characterized in that, The specific content of Step 1.4 is as follows: The power balance constraint is expressed as: in, Contribute to new energy forecasting For the predicted power of wind power, The predicted power output of photovoltaic power; This refers to the power of wind and solar power that has been curtailed. This refers to the discharge power of the energy storage system. Power for charging energy storage systems; Power generation capacity of distributed generating units; This value represents the power of the connection line with the main power grid. A positive value indicates that electricity is purchased from the grid, while a negative value indicates that electricity is sold to the grid. Base load power refers to the initial electricity demand of the park during the corresponding time period and scenario; Reduce power in response to demand; The dynamic constraint of the energy storage SOC is: in, This represents the state of charge of the energy storage system at the end of time period t+1 under scenario ω, and it represents the remaining energy in the energy storage device. This represents the state of charge of the energy storage system at the beginning of time period t under scenario ω, i.e., the initial energy state; This indicates the charging efficiency of the energy storage system. This indicates the discharge efficiency of the energy storage system; Indicates the unit time step of the scheduling; And satisfy: in: This indicates the minimum permissible state of charge for an energy storage system. This indicates the maximum allowable state of charge of the energy storage system, i.e., the rated capacity of the energy storage. This indicates the maximum allowable charging power of the energy storage system. This indicates the maximum allowable discharge power of the energy storage system.

5. The industrial park optimization scheduling method based on the wild horse herd algorithm according to claim 1, characterized in that, The specific content of Step 2 is as follows: Step 2.1: Respectively perform kernel density estimation on the three random variables of wind power output, photovoltaic power output and flexible load to obtain the corresponding marginal distribution functions; in, Let be the marginal distribution function of the j-th random variable, where j=1 represents wind power, j=2 represents photovoltaic power, and j=3 represents flexible load. The marginal probability density function is obtained through kernel density estimation; For kernel functions; It can be any one of the three random variables: wind power output, photovoltaic power output, or flexible load. Let be the sampled data points at time i for the j-th type of random variable; n is the number of sampled data points; h is the bandwidth, which controls the width of the kernel function; Each random variable is transformed into a uniformly distributed variable through uniform transformation. : in, To minimize the value and avoid boundary issues; Step 2.2: Use the tree structure of R-vine Copula to decompose the multi-dimensional dependence relationship, construct a joint probability distribution through a series of two-dimensional Copula functions, generate a large number of typical scenarios, and the joint probability density function is: in, , For Copula density function; Step 2.3: Adopt the K-means clustering algorithm for scenario reduction to reduce the computational complexity, and the objective function is: in, This represents the i-th original scene vector generated by sampling. It includes all data on wind power, solar power, and flexible loads over a period of time; This is the set of scenes contained in the k-th cluster; This represents the center vector of the k-th cluster, used to represent the typical scene features under that cluster; A set of typical scenarios was obtained through iterative optimization. This is used for subsequent optimization and scheduling, calculating the weight of each typical scenario. The calculation formula is as follows: in, This represents the number of original scenes assigned to the k-th cluster; N represents the total number of original scenes generated through sampling. Typical scenarios The weight.

6. The industrial park optimization scheduling method based on the wild horse herd algorithm according to claim 1, characterized in that, The specific content of Step 3 is as follows: Step 3.1: Take the typical scenario set generated by R-vine Copula and the corresponding weights as the uncertainty input, and establish an upper-layer robust optimization model with the minimum overall operating expected cost of the industrial park in Step 1 as the optimization goal; Step 3.2: Divide the park into several aggregation units, take the minimum operating cost of each unit as the local optimization objective, introduce consistency variables and use ADMM to carry out parallel distributed coordination iteration of each unit to obtain the current decision variables, including the charging and discharging power of the energy storage unit, the demand response reduction power of the flexible unit, and the output value of the distributed unit. Step 3.3: Determine whether the current output decision variables meet the convergence conditions. If the convergence conditions are met, the current charging and discharging power of the energy storage unit, the demand response reduction power of the flexible unit, and the output value of the distributed unit are output as the optimal solution. Step 3.4: If the convergence condition is not met, start the wild horse herd algorithm, map the decision variables to the positions of wild horse individuals, initialize the wild horse herd for each aggregation unit, and calculate the fitness value of each wild horse individual. Step 3.5: Select the wild horse individual with the smallest fitness value in each aggregation unit, return to step 3.2 for each aggregation unit to perform local iteration using the ADMM algorithm to obtain the next generation of decision variables, and determine whether the convergence condition is met. If it is met, output the next generation of decision variables; otherwise, continue to execute steps 3.4 and 3.5 until the convergence condition is met.

7. The industrial park optimization scheduling method based on the wild horse herd algorithm according to claim 6, characterized in that, Step 3.1 specifically involves: The park is divided into several aggregation units, and a set of decision variables is initialized for each unit. And introduce consistency variables ADMM iteration is used: in, Let be the local objective function, corresponding to the operating cost of the i-th schedulable unit; As a penalty factor; This is the coefficient matrix, used to... Mapped to a globally consistent space; Let k be a globally consistent variable, and k be the number of iterations. As a penalty factor, The dual variables reflect the deviation compensation between the local scheme and the global coordinated scheme; the current decision variables of each aggregation unit are obtained through the above formula.

8. The industrial park optimization scheduling method based on the wild horse herd algorithm according to claim 6, characterized in that, In step 3.3, convergence is determined according to the following formula: in, For the original residual, For dual residuals, Original residual tolerance, Dual residual tolerance.

9. The industrial park optimization scheduling method based on the wild horse herd algorithm according to claim 6, characterized in that, Step 3.4 specifically involves: Step 3.4.1: Initialize the wild horse population. Let the decision vector of the i-th aggregation unit be... The group size is M, and the position of the nth wild horse is... The fitness function is: in, Let be the fitness value of the nth wild horse individual in the i-th aggregation unit. The smaller the fitness value, the better the scheduling scheme. Let be the local objective function, corresponding to the operating cost of the i-th aggregation unit; To constrain violations and penalties; Step 3.4.2: The wild horse herd algorithm iteratively updates the individual positions until the maximum number of iterations or the fitness convergence threshold is reached, obtaining the optimal wild horse individual corresponding to each aggregation unit, i.e., the optimal decision variable.

Images