Mine integrated energy system optimal scheduling method based on multi-task differential evolution

Abstract

This invention discloses an optimal scheduling method for integrated mine energy systems based on multi-task differential evolution, belonging to the field of integrated mine energy scheduling technology. It includes constructing a multi-objective optimization model for the operation optimization problem of integrated mine energy systems; constructing a frequency-domain-assisted multi-task differential evolution algorithm framework, including a main task and auxiliary tasks. In the initialization phase, a main-auxiliary population initialization strategy based on Gaussian random fields is used to generate an initial population. In the evolution phase, an adaptive frequency-domain differential evolution strategy is used to continuously update the frequency-domain decision variables in the auxiliary tasks, outputting the optimal solution set. Through the collaborative solution of the main task and auxiliary tasks, the optimal operation scheme of the integrated mine energy system is obtained. This invention, by employing the above-mentioned optimal scheduling method for integrated mine energy systems based on multi-task differential evolution, effectively overcomes the limitations of traditional evolutionary methods in handling variables with temporal coupling relationships by collaboratively optimizing the time-domain main task and the frequency-domain auxiliary task.

Description

Technical Field

[0001] This invention relates to the field of integrated energy dispatching technology in mines, and in particular to an optimized dispatching method for integrated energy systems in mines based on multi-task differential evolution. Background Technology

[0002] Integrated Energy Systems (IES) can integrate multiple energy forms such as electricity, natural gas, and renewable energy to achieve coordinated planning, optimized operation, and collaborative management among various energy sources, and have become one of the important technologies for energy conservation and emission reduction in enterprises. Coal mine integrated energy systems (CMIES) are a typical yet unique type of IES system. While producing primary energy (coal), they also generate various other energy sources such as exhaust gas, air heat, and mine wastewater, and simultaneously require the timely utilization of renewable energy. The strong coupling between various energy sources and the strong constraints imposed by the actual scenario and equipment significantly increase the difficulty of solving the optimization problem of mine integrated energy systems.

[0003] Traditional methods for addressing the integrated energy system (CMIES) operation optimization problem often model it as a mixed-integer linear programming problem and employ mathematical programming solvers such as CPLEX and Gurobi. However, with ongoing research, existing CMIES operation optimization problems have increasingly exhibited complex characteristics such as large scale, nonlinearity, and multiple objectives, rendering traditional methods inadequate for practical applications. Due to their superior global search capabilities, evolutionary algorithms (EAs) have recently been used for CMIES operation optimization problems. However, constrained by various complex constraints, the feasible region of the CMIES operation optimization problem exhibits complex characteristics such as a small proportion and a discrete, irregular distribution, which severely limits the effectiveness of these algorithms.

[0004] The CMIES runtime optimization problem is essentially a class of constrained multi-objective optimization problems (CMOPs). Because the solution information from auxiliary tasks can guide the main task to quickly find the optimal solution to the original problem, and the difficulty of solving the auxiliary tasks is much lower than that of the original problem, multi-task evolutionary frameworks have become a popular technique for solving CMOPs. In recent years, scholars have also begun to try using the MTE framework to solve the CMIES runtime optimization problem. To guide the population to find the feasible region as quickly as possible, Ma et al. proposed a multi-task multi-objective evolutionary optimization algorithm that integrates evolutionary algorithms and mathematical programming. The main task uses evolutionary algorithms to solve the CMIES runtime optimization problem, while the auxiliary tasks use mathematical programming methods to solve a weighted single-objective optimization problem. Dai proposed a constraint-adaptive auxiliary task construction method within the multi-task framework, which can use the solution of infeasible regions to guide the algorithm to cross feasible regions. Furthermore, Dai et al. constructed multiple auxiliary tasks with different difficulty levels based on the processing difficulty of different types of constraints, and proposed an elite-guided knowledge transfer multi-task differential evolutionary algorithm. The above methods improve the performance of evolutionary algorithms in handling the CMIES runtime optimization problem. However, due to insufficient use of the knowledge of the constraints of the problem, they still suffer from a lack of diversity in the feasible solutions they obtain.

[0005] In addition to constraints related to equipment operation, the CMIES optimization problem also includes a special type of temporal constraint: the ramp-up constraint imposed by combined heat and power (CHP) units. This constraint limits the range of power output variation between adjacent time intervals, resulting in strong temporal coupling between different decision variables (especially adjacent variables). Specifically, a feasible solution to the problem exhibits smooth and continuous fluctuations in the values ​​of its corresponding variables. However, when using existing time-domain oriented coding algorithms (CMOEAs) to process the CMIES optimization problem, the algorithms cannot effectively capture these temporal coupling characteristics, which limits their performance to some extent.

[0006] Therefore, how to effectively handle the high-dimensionality and strong constraints (especially temporally coupled constraints) in the CMIES runtime optimization problem, and improve the algorithm's search efficiency and diversity for feasible solutions, has become a pressing technical challenge. Summary of the Invention

[0007] The purpose of this invention is to provide an optimized scheduling method for integrated energy systems in mines based on multi-task differential evolution, in order to solve the problems caused by the temporal coupling characteristics of variables and the high-dimensional characteristics of the time-domain search space.

[0008] To achieve the above objectives, this invention provides an optimized scheduling method for a mine integrated energy system based on multi-task differential evolution, comprising the following steps:

[0009] S1. Construct a multi-objective optimization model for the operation optimization problem of a comprehensive energy system in a mine;

[0010] S2. Construct a frequency-domain assisted multi-task differential evolution algorithm framework (FADE), which includes a main task and an auxiliary task. The main task solves the original problem in a high-dimensional time domain space, and the auxiliary task solves the original problem in a low-dimensional frequency domain space.

[0011] S21. Construct the main task using high-dimensional time-domain decision variables. Decision variables Divided equally Group, of which the first Subvariables For the first The device is The power over a period of time is used to solve the objective function using conventional differential evolution and constraint dominance principles to obtain a new population. The objective function is consistent with the multi-objective optimization model of S1.

[0012] S22. Construct auxiliary tasks to support the main task. The low-dimensional time-domain spectrum is obtained by performing Discrete Fourier Transform (DFT) on each of the group of time-domain subvariables. , Indicates Fourier transform, Spectrum Constitutes low-dimensional frequency domain decision variables The frequency domain differential evolution algorithm and its improvement are adopted. Constraint methods are used to solve the objective function and obtain a new population;

[0013] The objective function for the auxiliary task is:

[0014] (10)

[0015] in, and The frequency domain decision variables are respectively The corresponding operating cost function and energy curtailment cost function after conversion to the time domain. The total constraint violation value calculated by formulas (5)-(8) The maximum constraint violation value for all initial individuals. For the current generation, The maximum number of generations;

[0016] S23. Information exchange between the main task and the auxiliary task is achieved through time-domain to frequency-domain transformation. The frequency-domain solution of the auxiliary task is restored to the time domain by inverse discrete Fourier transform and then participates in the environment selection of the main task, guiding the main task's search. The high-quality time-domain solution of the main task is transformed to the frequency domain by DFT to guide the auxiliary task's search.

[0017] S3. In the initialization phase, a master-slave population initialization strategy based on Gaussian random fields is adopted to generate high-quality initial individuals with temporal continuity of variable values ​​in the master-slave population, thereby generating the initial population.

[0018] S4. During the evolutionary stage, an adaptive frequency domain differential evolution strategy is adopted to continuously update the frequency domain decision variables in the auxiliary task and output the optimal solution set, including a frequency domain differential mutation operator based on adaptive search weight and a time-frequency conversion mechanism.

[0019] S5. By solving the main task and auxiliary tasks in a coordinated manner, the optimal operation scheme of the integrated energy system of the mine is obtained.

[0020] Preferably, the specific steps of S1 are as follows:

[0021] S11. Define optimization objectives, including operating costs. and the cost of energy curtailment ;

[0022] The formula for calculating operating costs is:

[0023] (1)

[0024] (2)

[0025] (3)

[0026] in, The cost of purchasing traditional energy sources, For the cost of equipment operation and maintenance, Time period Internal power grid supply For electricity purchase price, This refers to the natural gas consumption of combined heat and power (CHP) equipment. For natural gas prices, For equipment The operating and maintenance cost coefficient, For equipment In time period Internal power output, A collection of electrical equipment. For absorption chiller units in a certain time period The internal cooling load consumption, and its corresponding cost is , This represents the total number of scheduling periods;

[0027] The formula for calculating the cost of energy curtailment is:

[0028] (4)

[0029] in, For equipment The penalty coefficient for wasting energy. For equipment In time period Maximum output power within, A collection of electrical equipment that involves energy curtailment;

[0030] S12. Determine the equipment operation constraints, including supply and demand balance constraints, energy conversion balance constraints, power upper and lower limit constraints, and ramping constraints;

[0031] S13. Construct a multi-objective optimization model for the operation optimization problem of the integrated energy system in a mine.

[0032] Preferably, the constraints for S12 are as follows:

[0033] a. The supply and demand balance constraint is:

[0034] (5)

[0035] in, and Solar and wind power, respectively, in time periods Internal power generation capacity For CHP units during the time period Internal power generation capacity Time period Internal electrical load, , , VOHP, WSHP, and EC are respectively in the time period Internal power consumption , , CHP, VOHP, and WSHP are respectively in the time period Internal heat output, Time period Internal heat load, For AC in the time period The internal heat power consumed, , EC and AC respectively in the time period Internal cold output, Time period Internal cooling load;

[0036] b. The energy conversion balance constraint is:

[0037] (6)

[0038] in, and The gas-to-heat and gas-to-electricity efficiency coefficients of the CHP unit. , , , These are the energy conversion coefficients for VOHP, WSHP, EC, and AC, respectively.

[0039] c. The upper and lower power limit constraints are:

[0040] (7)

[0041] in, and respectively equipment In time period The maximum and minimum values ​​of internal electrical power. For a collection of electrical devices involving power constraints, and respectively equipment In time period The maximum and minimum values ​​of internal cooling power. and The power of the grid in the time period The maximum and minimum values ​​within;

[0042] d. The climbing constraint is:

[0043] (8)

[0044] in, and These represent the maximum values ​​when the device's CHP power output increases and decreases, respectively.

[0045] Preferably, the specific expression of the model constructed by S13 is:

[0046] (9)

[0047] in, For time-domain decision variables The operating cost function, For time-domain decision variables The cost function of energy curtailment, decision variables Depend on Group independent variables composition, For the number of devices that need optimized scheduling, For the first The device is Power over a period of time.

[0048] Preferably, the specific steps of S3 are as follows:

[0049] S31. Set the parameters of the Gaussian random field and locate the expression of the Gaussian random field model as follows:

[0050] (11)

[0051] Each position Corresponding to a random variable , For random functions in a Gaussian random field model, For a continuous space, any finite set of points The function value at point follows a multivariate Gaussian distribution. , It is the mean vector. It is the covariance matrix;

[0052] S32, in continuous time Set up a set of equally spaced sampling points within the space , For set The first in One sampling point, This represents the total number of scheduling periods;

[0053] S33, Population Individuals k Sub-variables, for the first Sub-variables, sampled from Gaussian random fields A length of The sample sequence is normalized to the [0,1] interval, and linearly transformed to the device power range to obtain the first sample. sub-variables For each individual segment, repeat the above process until... All sub-variables are generated A fragment

[0054] S34, will sub-variables The samples are horizontally spliced ​​to generate the initial populations for the main task and auxiliary tasks.

[0055] Preferably, the specific steps of S4 are as follows:

[0056] S41. Initialize parameters, including population size. Maximum number of iterations and parameters used to control the search weights in the frequency domain differential mutation operator ;

[0057] S42, Update Parameters Mutant individuals are generated using a frequency-domain differential mutation operator based on adaptive search weights. Generating offspring individuals using the traditional binomial crossover operator ;

[0058] S43. Use a time-frequency conversion mechanism to convert frequency domain individuals into time domain individuals;

[0059] S44. When a component of an individual in the time domain is close to the boundary of the search space, a boundary repair strategy is executed, and the component value is taken as the boundary value closest to it.

[0060] S45. Repeat steps S42-S44 to output the optimal solution set for the comprehensive energy optimization scheduling problem in the mine.

[0061] Preferably, the frequency domain difference mutation operator based on adaptive search weights in S42 includes two mutation operators: DE / current-to-pbest / 1 and DE / current-to-rand / 1, where DE / current-to-pbest / 1 is the difference between the current individual and the previous individual in the population. The differential evolution mutation operator learned by % of random elite individuals, where DE / current-to-rand / 1 is the differential evolution mutation operator learned by the current individual and random individuals in the population:

[0062] Generate mutant individuals using DE / current-to-pbest / 1 The calculation formula is:

[0063] (12)

[0064] Generate mutant individuals using DE / current-to-rand / 1 The calculation formula is:

[0065] (13)

[0066] in, The first in the population Individual, The top of the current population % of random individuals, , , These are three distinct random individuals in the current population. For the search weight function, the dimensions and same;

[0067] No. Search weight function for each individual for:

[0068] (14)

[0069] in, It is the maximum number of generations the algorithm can evolve. This represents the current iteration number of the algorithm. For the weight vector, Contains K sets of identical weight vectors , The dimension of each weight vector, for The Middle The weight values ​​of each component, It is a function determined by the frequency of the components. A function controlled by the number of iterations. and These are parameters used to control the steepness of the slope;

[0070] parameter Updated using a Cauchy distribution obtained through random sampling, i.e. ;

[0071] Position parameters The update method is as follows:

[0072] (15)

[0073] in, For the first Generational population The parameter values ​​for each individual, For the first The Lehmer mean of all individuals in the population over a generation.

[0074] Preferably, the specific steps of the S43 time-frequency conversion mechanism are as follows:

[0075] S431. Transform from the time domain to the frequency domain, for time domain individuals of The discrete Fourier transforms are performed on the sub-variables to obtain the frequency domain representation of the time-domain individual;

[0076] For the first Group decision variables The discrete Fourier transform is expressed as follows:

[0077] (16)

[0078] in, For the first The original spectrum of the sub-variables, its dimension is , for The Middle Each variable can take a value. For the first in the spectrum One frequency component, For Fourier basis functions, The imaginary unit;

[0079] S432, Frequency domain individuals are Spectrum Composition, first reconstruct the complete spectrum, for Perform inverse Fourier transform on each complete spectrum to convert it into a single spectrum. Each spectrum is reconstructed to the time domain to obtain the time-domain representation of the frequency-domain decision variables;

[0080] The complete spectrum includes positive and negative frequency components, specifically represented as follows:

[0081] (17)

[0082] in, For conjugate operations, if It is an odd number, and the negative frequency component is composed of... The complex conjugate of the second to last components constitutes the structure; if If it is even, then take The second to second-to-last components are conjugated.

[0083] The formula for the inverse Fourier transform is:

[0084] (18)

[0085] in, These are the basis functions of the inverse Fourier transform.

[0086] Preferably, the specific formula for the S44 boundary repair strategy is as follows:

[0087] (19)

[0088] in, For the first Individual, the first The values ​​of each decision component after boundary repair The upper boundary, This is the lower boundary.

[0089] Therefore, the above-mentioned optimization scheduling method for integrated energy systems in mines based on multi-task differential evolution has the following beneficial effects:

[0090] (1) A frequency-domain-assisted multi-task evolutionary optimization framework is proposed. The original problem (i.e., the main task) is solved in the time-domain variable space, and the variable with mutation behavior is optimized by utilizing the characteristic that the time-domain signal is suitable for representing transient behavior. The auxiliary task is constructed and solved in the frequency-domain variable space. While ensuring that the obtained solution is more likely to satisfy the temporal constraints, the search space of the population is significantly reduced. The two tasks are solved in a multi-task framework, which can significantly improve the algorithm's ability to solve the CMIES problem.

[0091] (2) An adaptive frequency domain differential evolution algorithm is proposed to adapt to the optimization of frequency domain decision variables in auxiliary tasks. A frequency domain differential mutation operator based on adaptive search weight is proposed so that the algorithm focuses on optimizing variables of different frequencies at different evolution stages, thereby effectively balancing the global search and local development capabilities of the algorithm in the frequency domain. At the same time, a time domain-frequency domain individual conversion mechanism is designed to obtain the correct representation of the individual in the frequency domain and effectively evaluate the fitness value of the individual in the frequency domain.

[0092] (3) A master-slave population initialization method based on Gaussian random fields is proposed. Gaussian random fields can describe the continuous change characteristics between different variables. By sampling Gaussian random fields, initial individuals with continuous time relationships can be obtained, thereby improving the quality of initial individuals.

[0093] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0094] Figure 1 This is a flowchart of an optimized scheduling method for a mine integrated energy system based on multi-task differential evolution, according to the present invention.

[0095] Figure 2 This is a basic architecture diagram of a mine integrated energy system, which is based on a multi-task differential evolution optimization scheduling method of the present invention.

[0096] Figure 3 This is a basic framework diagram of the FAMDE (Multi-Task Differential Evolution) method for optimizing the scheduling of integrated energy systems in mines, as proposed in this invention.

[0097] Figure 4 The weight vector of this invention is a multi-task differential evolution-based optimization scheduling method for integrated energy systems in mines. picture;

[0098] Figure 5 This invention presents a method for optimizing the scheduling of a comprehensive energy system in a mine based on multi-task differential evolution, which includes a power allocation and consumption balance diagram for electricity (E-.), heat (H-.), and cold (C-.). Detailed Implementation

[0099] The following detailed description of 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.

[0100] Example

[0101] like Figure 1 As shown, this invention provides an optimized scheduling method for a mine integrated energy system based on multi-task differential evolution, comprising the following steps:

[0102] S1. Construct a multi-objective optimization model for the operation optimization problem of the integrated energy system in a mine.

[0103] The Integrated Mining Energy System (CMIES) consists of three core components: energy supply, energy conversion, and energy consumption. On the energy supply side, the system integrates various energy forms, including traditional energy (natural gas and grid power), renewable energy (photovoltaic (PV) and wind turbine (WT)), and mine-derived energy (ventilation air methane (VAM), air heat, mine wastewater, and geothermal energy, etc.). In the energy conversion stage, the system is equipped with various advanced energy conversion devices, including combined heat and power (CHP) units, VAM oxidation heat pumps (VOHP), water-source heat pumps (WSHP), absorption chillers (AC), and electric chillers (EC). These devices work together to efficiently convert different forms of energy into the energy forms required by end users. On the energy demand side, the system primarily meets the diverse energy needs of users, including electricity load, cooling load, and heating load.

[0104] CMIES infrastructure such as Figure 2 As shown; this embodiment addresses the day-ahead scheduling optimization problem of CMIES, considering two optimization objectives: system operating cost and the cost of abandoned renewable energy and associated coal mine energy (referred to as abandoned energy cost), and constructs a multi-objective optimization model for the CMIES operation optimization problem.

[0105] S11. Define optimization objectives, including operating costs. and the cost of energy curtailment ;

[0106] The formula for calculating operating costs is:

[0107] (1)

[0108] (2)

[0109] (3)

[0110] in, The cost of purchasing traditional energy sources, For the cost of equipment operation and maintenance, Time period Internal power grid supply For electricity purchase price, This refers to the natural gas consumption of combined heat and power (CHP) equipment. For natural gas prices, For equipment The operating and maintenance cost coefficient, For equipment In time period Internal power output, The term refers to a collection of electrical equipment, which includes wind turbines (WT), photovoltaic power generation (PV), combined heat and power (CHP), volatile organic compound (VAM) oxidation heat pumps (VOHP), water source heat pumps (WSHP), and electric chillers (EC). For absorption chillers (AC) during the time period The internal cooling load consumption, and its corresponding cost is ;

[0111] The cost of energy curtailment mainly includes the cost of discarding renewable energy and associated energy from coal mines, and the calculation formula is as follows:

[0112] (4)

[0113] in, For equipment The penalty coefficient for wasting energy. For equipment In time period Maximum output power within, This refers to a collection of electrical equipment involving energy curtailment, including wind turbines (WT), photovoltaic (PV), vapor-oxidizing heat pumps (VOHP), and water source heat pumps (WSHP).

[0114] S12. Determine the equipment operation constraints, including supply and demand balance constraints, energy conversion balance constraints, power upper and lower limit constraints, and ramping constraints.

[0115] Stable system operation requires strict adherence to various constraints, including equation constraints reflecting supply and demand balance and energy conversion balance, inequality constraints limiting the operating range of equipment, and ramp constraints (timing constraints) limiting changes in equipment output power.

[0116] a. Supply and demand balance constraint, describing the equationual relationship between the output power and power consumption of all devices during system operation, with the following condition:

[0117] (5)

[0118] in, and Photovoltaics, In time period Internal power generation capacity For CHP units during the time period Internal power generation capacity Time period Internal electrical load, , , VOHP, WSHP, and EC are respectively in the time period Internal power consumption , , CHP, VOHP, and WSHP are respectively in the time period Internal heat output, Time period Internal heat load, For AC in the time period The internal heat power consumed, , EC and AC respectively in the time period Internal cold output, Time period The internal cooling load.

[0119] b. Energy conversion balance constraints describe the balance relationship between different types of energy (such as electricity, heat, and cold) in energy conversion, under the following conditions:

[0120] (6)

[0121] in, and The gas-to-heat and gas-to-electricity efficiency coefficients of the CHP unit. , , , These are the energy conversion coefficients for VOHP, WSHP, EC, and AC, respectively.

[0122] c. Power upper and lower limit constraints, describing the normal power range of the equipment, with the following conditions:

[0123] (7)

[0124] in, and respectively equipment In time period The maximum and minimum values ​​of internal electrical power. For a collection of electrical devices involving power constraints, and respectively equipment In time period The maximum and minimum values ​​of internal cooling power. and The power of the grid in the time period The maximum and minimum values ​​within.

[0125] d. Climbing constraint: A combined heat and power (CHP) unit consumes natural gas and simultaneously outputs heat and electricity. The electrical output power of this unit is limited by a climbing constraint, meaning the degree of variation in its output power between adjacent time periods is restricted. The condition is as follows:

[0126] (8)

[0127] in, and These represent the maximum values ​​when the device's CHP power output increases and decreases, respectively.

[0128] S13. Construct a multi-objective optimization model for the operation optimization problem of a mine's integrated energy system.

[0129] The aforementioned CMIES problem requires optimizing the power of eight devices across three energy forms over a 24-hour period. These devices include PV, WT, CHP, VOHP, WSHP, AC, EC, and grid. Since the cooling and heating power of these devices can be obtained through energy conversion relationships (equations), only the electrical power of these devices needs to be optimized. Furthermore, by applying variable elimination to the equations (equations), the output power of CHP, AC, and grid devices can be eliminated. Therefore, only the power of the remaining five devices needs to be optimized. Considering a system scheduling interval of one hour (i.e., the number of system scheduling periods), each device contains 24 power values ​​that need optimization. Thus, the CMIES day-ahead scheduling problem to be optimized in this embodiment contains 120 decision variables and 95 inequality constraints, the specific expressions of which are as follows:

[0130] (9)

[0131] in, For time-domain decision variables The operating cost function, For time-domain decision variables The cost function of energy curtailment, decision variables Depend on Group independent variables composition, For the number of devices that need optimized scheduling, For the first The device is Power over a period of time.

[0132] In the above model, the inequality constraints include , , The CMIES day-ahead scheduling optimization problem involves constraints on the working interval and the CHP ramping constraint. The ramping constraint, as a time-series constraint, requires that the values ​​of adjacent variables exhibit a smooth change. Furthermore, the ramping constraint is also a dynamic decision-space constraint, requiring that the constraint conditions are related to the variable values, thus forming a dynamic decision space. It is worth noting that although the ramping constraint directly affects the CHP unit, its impact is transmitted to other equipment through the system's supply-demand balance, thereby affecting the range of variable changes in adjacent time periods for other equipment. Therefore, the CMIES day-ahead scheduling optimization problem exhibits complex characteristics such as high dimensionality, strong constraints, strong variable coupling, and dynamic decision-space constraints.

[0133] S2. Construct a frequency-domain-assisted multi-task differential evolution algorithm framework, including a main task and an auxiliary task. The main task solves the original problem in a high-dimensional time domain space, and the auxiliary task solves the original problem in a low-dimensional frequency domain space.

[0134] The basic framework of the FAMDE algorithm is as follows: Figure 3 As shown, a low-dimensional auxiliary task is constructed in the frequency domain variable space to solve the CMIES runtime optimization problem with temporally coupled constraints. Compared with existing multi-task evolutionary optimization frameworks, this framework can effectively integrate the advantages of time-domain and frequency-domain variable representations. Specifically, decision variables represented in the time domain are suitable for optimizing variables with significant discontinuities between adjacent time periods, while decision variables represented in the frequency domain quickly search for feasible solutions with temporally continuous characteristics by optimizing the frequency components of individuals throughout the entire time period. For example, the time-domain main task can directly optimize the grid connection status of photovoltaic devices in CMIES, where the power values ​​at adjacent time points exhibit step changes. The frequency-domain auxiliary task can optimize the long-term trend and local fluctuations of the 24-hour power of CHP devices by optimizing the low-frequency and high-frequency components in the frequency-domain decision variables, thereby quickly searching for solutions that satisfy the temporal constraints.

[0135] Although the primary and auxiliary tasks of FAMDE solve the same problem, their decision spaces, evolutionary strategies, and environment selection strategies differ. The primary task solves the primal problem in a high-dimensional time domain using conventional differential evolution (DE) and the constraint dominance principle (CDP), while the auxiliary task solves the problem in a low-dimensional frequency domain using frequency-domain differential evolution and improved... The constraint method is used to solve the original problem. To extend differential evolution to the frequency domain variable space, new strategies are proposed, such as the master-slave population initialization strategy based on Gaussian random fields and frequency domain differential mutation with adaptive search weights.

[0136] The pseudocode for FAMDE is shown in Algorithm 1. It should be noted that: (1) the auxiliary population is initialized using a time-domain representation (line 2), and in Algorithm 3, a frequency-domain to time-domain individual conversion mechanism is performed beforehand to convert the frequency-domain individuals to the time domain in order to evaluate the individuals in the auxiliary population (lines 4 and 8); (2) the main task can use any traditional DE mutation operator, specifically DE / current-to-pbest / 1 and DE / current-to-rand / 1; (3) initial parameters need to be set for individuals in both the main and auxiliary tasks. However, only the individual parameters in the auxiliary task need to be updated. (4) In the environment selection phase, information interaction between the main and auxiliary tasks is realized, which can effectively handle the situation where the optimal feasible region is far away from the unconstrained Pareto front. Specifically, the two offspring are first merged into the initial populations of the main task and the auxiliary task respectively. Then, the main task uses the CDP method to obtain a new population, while the auxiliary task uses the improved method. Methods to obtain new populations.

[0137]

[0138] S21. Construct the main task using high-dimensional time-domain decision variables. Due to the problems with CMIES, including One device to be optimized, decision variables Divided equally Group, of which the first Subvariables For the first The device is The output situation within a certain time period, in other words, the output situation within the first time period. Group sub-variables include The first component, its second component The component represents its value in the th... The equipment output at any given time means that only the components within a group have temporal correlation; a new population is obtained by solving the objective function using conventional differential evolution and constraint dominance principles, and the objective function is consistent with the multi-objective optimization model of S1;

[0139] S22. Based on the characteristics of CMIES problems, construct auxiliary tasks to support the main task. Each group of time-domain sub-variables undergoes a Discrete Fourier Transform (DFT) to convert the time-domain decision variables of each group into their corresponding frequency-domain representation (i.e., the spectrum represented by complex numbers). A frequency-domain decision variable... Depend on A low-dimensional time-domain spectrum Composition, each spectrum has a dimension of That is, the spectrum includes One frequency component, Representing the Fourier transform, using frequency domain differential evolution and improvement. Constraints are applied to solve the objective function to obtain a new population;

[0140] The objective function for the auxiliary task is:

[0141] (10)

[0142] in, and The frequency domain decision variables are respectively The corresponding operating cost function and energy curtailment cost function after conversion to the time domain. The total constraint violation value calculated by formulas (5)-(8) The maximum constraint violation value for all initial individuals. For the current generation, This represents the maximum number of generations; it should be noted that when calculating a set of frequency variables... Two target values and At this time, it is necessary to first execute the frequency domain to time domain individual conversion mechanism to convert the individual in the frequency domain to the real number domain.

[0143] S23. Information exchange between the main task and the auxiliary task is achieved through time-domain to frequency-domain transformation. The frequency-domain solution of the auxiliary task is restored to the time domain by inverse discrete Fourier transform and then participates in the environment selection of the main task, assisting the main task in its search. The high-quality time-domain solution of the main task is transformed to the frequency domain by DFT to guide the search of the auxiliary task.

[0144] S3. In the initialization phase, a master-slave population initialization strategy based on Gaussian random fields is adopted to generate high-quality initial individuals with temporal continuity of variable values ​​in the master-slave population, thereby generating the initial population.

[0145] In the comprehensive energy operation optimization problem, due to the climbing constraint, the values ​​of two adjacent variables within a feasible candidate solution (individual) should not vary too much, i.e., adjacent variables have temporal coupling characteristics. Traditional population initialization methods (such as uniform distribution) usually assume that different variables are independent of each other, which leads to the generated initial individuals violating the above temporal constraints to a greater extent. Therefore, this implementation proposes a master-slave population initialization strategy based on Gaussian Random Field (GRF).

[0146] S31. A Gross Random Field (GRF) is a statistical model describing the temporal or spatial variation of a random phenomenon, where the changes between adjacent variables in the field are continuous. Specifically, a GRF is a stochastic process defined in a continuous space (such as time, geographical region, etc.), and the joint distribution of any finite set of points follows a multivariate Gaussian distribution. The expression for defining the GRF by setting its parameters is as follows:

[0147] (11)

[0148] Each position Corresponding to a random variable , For random functions in a Gaussian random field model, For a continuous space, any finite set of points The function value at point follows a multivariate Gaussian distribution. , It is the mean vector. For the covariance matrix; set , ,in It is the Lap operator. For unit operators; based on the Karhunen-Loeve expansion method, by setting the inclusion... The point set of sampling points A A set of sample sequences with continuous time characteristics can be generated by sampling from a Gaussian random field, denoted as GRF (Gaussian Random Field). A ).

[0149] S32, in continuous time Set up a set of equally spaced sampling points within the space , For set The first in One sampling point, This represents the total number of scheduling periods;

[0150] S33, Population Individuals k Sub-variables, for the first Sub-variables, sampled from Gaussian random fields A length of The sample sequence is normalized to the [0,1] interval, and linearly transformed to the device power range to obtain... sub-variables For each individual segment, repeat the above process until... All sub-variables are generated A fragment

[0151] S34, will sub-variables The samples are horizontally spliced ​​to generate the initial populations for the main task and auxiliary tasks.

[0152] Specifically, it is expressed as follows:

[0153]

[0154] S4. In the evolutionary stage, in order to effectively update the frequency domain decision variables, an adaptive frequency domain differential evolution strategy is adopted to continuously update the frequency domain decision variables in the auxiliary task and output the optimal solution set, including a frequency domain differential mutation operator based on adaptive search weights and a time-frequency conversion mechanism.

[0155] Compared to existing time-domain DE algorithms, the adaptive frequency-domain differential evolution strategy introduces two new operators: a frequency-domain differential mutation operator with adaptive search weights and a time-frequency transformation mechanism. The frequency-domain differential evolution mutation operator is used to generate new individuals, while the time-frequency transformation mechanism is used to convert between frequency-domain and time-domain individuals.

[0156] S41. Initialize parameters, including population size. Maximum number of iterations and parameters used to control the search weights in the frequency domain differential mutation operator .

[0157] S42, Update Parameters Mutant individuals are generated using a frequency-domain differential mutation operator based on adaptive search weights. Generating offspring individuals using the traditional binomial crossover operator .

[0158] In the frequency domain, the low-frequency components of the spectrum determine the overall trend of the time-domain signal, while the high-frequency components reflect its local details. The scaling factor F in the traditional DE mutation operator can only scale the difference vector as a whole, and cannot independently adjust each of its components. When used to optimize frequency-domain decision variables, this adjustment method cannot distinguish the differentiated effects of different frequency components on global and local features, thus limiting the algorithm's search performance.

[0159] This embodiment proposes a frequency domain differential mutation operator based on adaptive search weights, using a set of search weights. To control the degree of mutation of the algorithm for different frequency components; specifically, to balance the diversity of the algorithm during global search and the convergence of the algorithm during local search, the two commonly used mutation operators DE / current-to-pbest / 1 and DE / current-to-rand / 1 are improved to obtain the following two frequency domain difference mutation operators: including DE / current-to-pbest / 1 and DE / current-to-rand / 1, where DE / current-to-pbest / 1 is the mutation operator from the current individual to the previous individual. The differential evolution mutation operator learned by % of random elite individuals, where DE / current-to-rand / 1 is the differential evolution mutation operator learned from the current individual to random individuals:

[0160] Generate mutant individuals using DE / current-to-pbest / 1 The calculation formula is:

[0161] (12)

[0162] Generate mutant individuals using DE / current-to-rand / 1 The calculation formula is:

[0163] (13)

[0164] in, The first in the population Individual, The top of the current population % of random individuals, , , These are three distinct random individuals in the current population. For the search weight function, the dimensions and same;

[0165] The purpose of introducing search weights is to dynamically adjust the search weight values ​​of different frequency components, so that the population focuses on developing low-frequency components in the early stages of evolution to improve the algorithm's global search capability; and in the later stages of evolution, it focuses on developing high-frequency components to achieve a fine-grained local search of potential regions. Based on spectral characteristics, low-frequency components are mainly distributed in the front part of the spectrum, while high-frequency components are mainly concentrated in the back part. Based on this, the... Search weight function for each individual for:

[0166] (14)

[0167] in, It is the maximum number of generations the algorithm can evolve. This represents the current iteration number of the algorithm. For the weight vector, Include K The same weight vector , The dimension of each weight vector, for The Middle The weight of each component, This is a function determined by the component frequencies, and its value determines the degree to which the algorithm favors different components at the same time. This function... The values ​​at each point are represented as discrete points on the Sigmoid curve, introducing... The goal is to increase the weight of low-frequency components and suppress the weight of high-frequency components; conversely, This emphasizes high-frequency components while suppressing low-frequency components. This is a function controlled by the number of iterations; its value determines the degree to which the algorithm favors one of the two emphases at different iteration stages. and The parameter used to control the steepness, initial value of the parameter. When the value is set to 0.5, approximately 50% of the frequencies receive a higher weight, while the remaining components are assigned a lower weight.

[0168] parameter The inflection point of the Sigmoid curve can be adjusted, and its value controls the number of frequency components that are the main optimization parameters. Theoretically, the algorithm can achieve the fastest convergence speed when all individuals have obtained the optimal parameter configuration, but this condition is prone to premature convergence of the population. To balance the convergence and diversity of the algorithm, Cauchy distribution is updated by random sampling. ,Right now ;

[0169] Position parameters The update method is as follows:

[0170] (15)

[0171] in, For the first The era The parameter values ​​for each individual, For the first The Lehmer mean of all individuals at that time; the above update method uses both individual historical values ​​and the population mean to adjust the Cauchy distribution, while maintaining... Under the premise of diversity, it can be gradually converged, thereby achieving the goal of balancing the convergence and diversity of the population.

[0172] In parameter configuration , , Under the conditions, Figure 4 Showing the weight vector The dynamic adjustment patterns that occur during evolution. The points on the red / white curves illustrate this. The weight changes of different frequency components before and after evolution are shown. It can be seen that in the early stages of evolution (using the red curve as an example), low-frequency components received higher search weights, while high-frequency components had lower weights. As the number of iterations increases, The trend gradually shifted to one where low-frequency components had lower search weights, while high-frequency components had higher weights, as shown by the white curve. Here, the parameters... By controlling the position of the gray plane, the timing of switching between higher and lower weights can be adjusted. By setting search weights in this way, the auxiliary task can gradually transition from primarily focusing on searching low-frequency components to prioritizing searching high-frequency components, smoothly shifting from global search to local search, thereby effectively balancing the algorithm's global and local search capabilities.

[0173] S43. Use a time-frequency conversion mechanism to convert frequency domain individuals into time domain individuals;

[0174] The specific steps of the time-frequency conversion mechanism are as follows:

[0175] S431. Transform from the time domain to the frequency domain, for time domain individuals of The discrete Fourier transforms are performed on the sub-variables to obtain the frequency domain representation of the time-domain individual;

[0176] For the first Group decision variables The discrete Fourier transform is expressed as follows:

[0177] (16)

[0178] in, For the first The original spectrum of the sub-variables, its dimension is , for The Middle Each variable can take a value. For the first in the spectrum One frequency component, These are Fourier basis functions used to project time-domain variables to... One frequency component, The imaginary unit;

[0179] S432, Frequency domain individuals are Spectrum Composition, to make this Each spectrum is then reconstructed to the time domain to obtain the time-domain representation of the frequency-domain decision variables; it is necessary to first reconstruct the complete spectrum, and then... Perform inverse Fourier transform on each complete spectrum, i.e. , will this Each spectrum is reconstructed to the time domain to obtain the time-domain representation of the frequency domain decision variables; the complete spectrum includes positive and negative frequency components, and the positive frequency is represented by the conjugate symmetry property. Partially expanded to the full spectrum Specifically, it is expressed as:

[0180] (17)

[0181] in, For conjugate operation; when the first The time-domain signal length corresponding to the group spectrum When it is an odd number, its negative frequency part is composed of The complex conjugate of the second component to the last component constitutes the structure. If If it is even, then take The second to second-to-last frequency components are processed using conjugate symmetry, and the formula for the inverse Fourier transform is:

[0182] (18)

[0183] in, These are the basis functions of the inverse Fourier transform.

[0184] S44. Considering that the optimal Pareto solution set of the integrated energy operation problem often falls on the boundary of the search space, when a certain component of an individual in the time domain is small from the boundary of the search space, a boundary repair strategy is executed, and the value of the component is taken from the boundary value closest to it.

[0185] The specific formula for the boundary repair strategy is as follows:

[0186] (19)

[0187] in, For the first Individual, the first The values ​​of each decision component after boundary repair The upper boundary, This is the lower boundary.

[0188] S45. Repeat steps S42-S44 to output the optimal solution set for the comprehensive energy optimization scheduling problem in the mine.

[0189] Specifically, it is expressed as follows:

[0190]

[0191] S5. By solving the main task and auxiliary tasks in a coordinated manner, the optimal operation scheme of the integrated energy system of the mine is obtained.

[0192] To verify the effectiveness of the Pareto optimal scheduling scheme obtained by the proposed algorithm, this embodiment randomly selects a scheduling scheme for analysis. Figure 5 The display shows the power distribution and demand balance of eight devices under three energy forms: electricity, heat, and cooling. The power input to the system is positive, the power consumption is negative, and Load represents the user load.

[0193] from Figure 5 As can be seen, in terms of power dispatch, the system can prioritize the consumption of renewable energy sources such as wind power during nighttime periods without solar power, and optimize electricity costs during daytime high-price periods (8:00-15:00) by reducing grid purchases, activating CHP units (consuming natural gas) with stable prices, and maximizing the utilization of wind and solar renewable energy. In terms of thermal dispatch, the system mainly relies on two derivative energy devices, VOHP and WSHP, reducing the output ratio of CHP units and demonstrating the environmentally friendly characteristics of the solution. In terms of cooling dispatch, the system uses thermally driven AC instead of electrically driven EC for cooling during high-price periods, resulting in significant economic benefits. In summary, this embodiment effectively coordinates the dispatch of various energy sources, ensuring a balance between energy supply and demand while effectively achieving the dual optimization goals of economic benefits and environmental protection, fully validating the practical value of the proposed algorithm.

[0194] Furthermore, experimental research on the integrated energy system of a mine in Shaanxi, China, in this embodiment shows that the proposed FAMDE algorithm can obtain a set of Pareto optimal solutions with excellent convergence and diversity, i.e., the optimal scheduling scheme.

[0195] To comprehensively evaluate the performance of the FAMDE algorithm, this embodiment selects 11 advanced algorithms across 5 categories as benchmark comparison algorithms, specifically including: decomposition: CMOEAD; multi-stage: MSCMO, Top; multi-task: CMOPT, CA-MTDE, IMTCM, EMCMO, CMOQLMT; co-evolution: CCMO, cDPEA; and metric selection: ICMA. All algorithms are implemented using default parameters on the platEMO platform, with a uniform maximum of 300,000 evaluation runs and 30 independent runs. The performance of the algorithms is evaluated using three metrics: hypervolume (HV), spread, and run time. A higher HV value indicates better solution set diversity and convergence; a lower spread value indicates better solution set diversity. Furthermore, the Wilcoxon rank-sum test at a p=0.05 significance level is used to evaluate the significant superiority of an algorithm. The symbols "+", "-", and "=" indicate that the performance of the comparison algorithm is significantly better than, worse than, or approximately equal to the proposed algorithm, respectively. Table 1 shows the statistical results of FAMDE and 11 comparison algorithms on three key performance indicators. All results are presented in the form of "mean (variance)", with the best value marked in bold.

[0196] Table 1. Statistical results of FAMDE and 11 comparison algorithms on three performance metrics.

[0197]

[0198] As shown in Table 1, the FAMDE algorithm significantly outperforms all the comparison algorithms in three key metrics: HV, Spread, and runtime. Specifically, in terms of convergence and stability, FAMDE not only achieves the best mean HV (1.19e-3) but also has the smallest variance, indicating excellent stability. Secondly, in terms of solution diversity, FAMDE's Spread metric is 1.5-3 times better than the comparison algorithms, and its optimal solution set exhibits excellent uniformity. Finally, in terms of computational efficiency, except for the Top and CA-MTDE algorithms, FAMDE's runtime is 2-3 times shorter than the comparison algorithms. These results fully demonstrate that the FAMDE algorithm maintains excellent convergence and diversity while also possessing high computational efficiency, exhibiting significant comprehensive advantages.

[0199] Therefore, the present invention adopts the above-mentioned optimization scheduling method for integrated energy systems in mines based on multi-task differential evolution. By coordinating the optimization of the time-domain main task and the frequency-domain auxiliary task, it effectively overcomes the limitations of traditional methods in dealing with variables with temporal coupling relationships.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for optimizing the scheduling of a mine integrated energy system based on multi-task differential evolution, characterized in that, Includes the following steps: S1. Construct a multi-objective optimization model for the operation optimization problem of a comprehensive energy system in a mine; S2. Construct a frequency-domain-assisted multi-task differential evolution algorithm framework, including a main task and an auxiliary task. The main task solves the original problem in a high-dimensional time domain space, and the auxiliary task solves the original problem in a low-dimensional frequency domain space. S21. Construct the main task using high-dimensional time-domain decision variables. Decision variables Divided equally Group, of which the first Subvariables For the first The device is The power over a period of time is used to solve the objective function using conventional differential evolution and constraint dominance principles to obtain a new population. The objective function is consistent with the multi-objective optimization model of S1. S22. Construct auxiliary tasks to support the main task. The low-dimensional time-domain spectrum is obtained by performing Discrete Fourier Transform (DFT) on each of the group of time-domain subvariables. , Indicates Fourier transform, Spectrum Constitutes low-dimensional frequency domain decision variables The frequency domain differential evolution algorithm and its improvement are adopted. Constraint methods are used to solve the objective function and obtain a new population; The objective function for the auxiliary task is: （10） in, and The frequency domain decision variables are respectively The corresponding operating cost function and energy curtailment cost function after conversion to the time domain. The total constraint violation value calculated by formulas (5)-(8) The maximum constraint violation value for all initial individuals. For the current generation, The maximum number of generations; S23. Information interaction between the main task and the auxiliary task is realized through time-domain to frequency-domain conversion. The frequency domain solution of the auxiliary task is restored to the time domain by inverse discrete Fourier transform and then participates in the environment selection of the main task to guide the main task search. The high-quality time domain solution of the main task is converted to the frequency domain by DFT to guide the auxiliary task search. S3. In the initialization phase, a master-slave population initialization strategy based on Gaussian random fields is adopted to generate high-quality initial individuals with temporal continuity of variable values ​​in the master-slave population, thereby generating the initial population. S4. During the evolutionary stage, an adaptive frequency domain differential evolution strategy is adopted to continuously update the frequency domain decision variables in the auxiliary task and output the optimal solution set, including a frequency domain differential mutation operator based on adaptive search weight and a time-frequency conversion mechanism. S5. By solving the main task and auxiliary tasks in a coordinated manner, the optimal operation scheme of the integrated energy system of the mine is obtained.

2. The optimal scheduling method for a mine integrated energy system based on multi-task differential evolution as described in claim 1, characterized in that, The specific steps of S1 are as follows: S11. Define optimization objectives, including operating costs. and the cost of energy curtailment ; The formula for calculating operating costs is: （1） （2） （3） in, The cost of purchasing traditional energy sources, For the cost of equipment operation and maintenance, Time period Internal power grid supply For electricity purchase price, This refers to the natural gas consumption of combined heat and power (CHP) equipment. For natural gas prices, For equipment The operating and maintenance cost coefficient, For equipment In time period Internal power output, A collection of electrical equipment. For absorption chiller units in a certain time period The internal cooling load consumption, and its corresponding cost is , This represents the total number of scheduling periods; The formula for calculating the cost of energy curtailment is: （4） in, For equipment The penalty coefficient for wasting energy. For equipment In time period Maximum output power within, A collection of electrical equipment that involves energy curtailment; S12. Determine the equipment operation constraints, including supply and demand balance constraints, energy conversion balance constraints, power upper and lower limit constraints, and ramping constraints; S13. Construct a multi-objective optimization model for the operation optimization problem of the integrated energy system in a mine.

3. The optimal scheduling method for a mine integrated energy system based on multi-task differential evolution as described in claim 2, characterized in that, The constraints of S12 are as follows: a. The supply and demand balance constraint is: （5） in, and Photovoltaics, In time period Internal power generation capacity For CHP units during the time period Internal power generation capacity Time period Internal electrical load, , , VOHP, WSHP, and EC are respectively in the time period Internal power consumption , , CHP, VOHP, and WSHP are respectively in the time period Internal heat output, Time period Internal heat load, For AC in the time period The internal heat power consumed, , EC and AC respectively in the time period Internal cold output, Time period Internal cooling load; b. The energy conversion balance constraint is: （6） in, and The gas-to-heat and gas-to-electricity efficiency coefficients of the CHP unit. , , , These are the energy conversion coefficients for VOHP, WSHP, EC, and AC, respectively. c. The upper and lower power limit constraints are: （7） in, and respectively equipment In time period The maximum and minimum values ​​of internal electrical power. For a collection of electrical devices involving power constraints, and respectively equipment In time period The maximum and minimum values ​​of internal cooling power. and The power of the grid in the time period The maximum and minimum values ​​within; d. The climbing constraint is: （8） in, and These represent the maximum values ​​when the device's CHP power output increases and decreases, respectively.

4. The optimal scheduling method for a mine integrated energy system based on multi-task differential evolution as described in claim 3, characterized in that, The specific expression of the model constructed by S13 is: （9） in, For time-domain decision variables The operating cost function, For time-domain decision variables The cost function of energy curtailment, decision variables Depend on Group independent variables composition, For the number of devices that need optimized scheduling, For the first The device is Power over a period of time.

5. The optimal scheduling method for a mine integrated energy system based on multi-task differential evolution as described in claim 1, characterized in that, The specific steps for S3 are as follows: S31. Set the parameters of the Gaussian random field and locate the expression of the Gaussian random field model as follows: （11） Each position Corresponding to a random variable , For random functions in a Gaussian random field model, For a continuous space, any finite set of points The function value at point follows a multivariate Gaussian distribution. , It is the mean vector. It is the covariance matrix; S32, in continuous time Set up a set of equally spaced sampling points within the space , For set The first in One sampling point, This represents the total number of scheduling periods; S33, Population Individuals k Sub-variables, for the first Sub-variables, sampled from Gaussian random fields A length of The sample sequence is normalized to the [0,1] interval, and linearly transformed to the device power range to obtain the first sample. Subvariables For each individual segment, repeat the above process until... All sub-variables are generated A fragment S34, will Subvariables The samples are horizontally spliced ​​to generate the initial populations for the main task and auxiliary tasks.

6. The optimal scheduling method for a mine integrated energy system based on multi-task differential evolution according to claim 1, characterized in that, The specific steps for S4 are as follows: S41. Initialize parameters, including population size. Maximum number of iterations and parameters used to control the search weights in the frequency domain differential mutation operator ; S42, Update Parameters Mutant individuals are generated using a frequency-domain differential mutation operator based on adaptive search weights. Generating offspring individuals using the traditional binomial crossover operator ; S43. Use a time-frequency conversion mechanism to convert frequency domain individuals into time domain individuals; S44. When a component of an individual in the time domain is close to the boundary of the search space, a boundary repair strategy is executed, and the component value is taken as the boundary value closest to it. S45. Repeat steps S42-S44 to output the optimal solution set for the comprehensive energy optimization scheduling problem in the mine.

7. The optimal scheduling method for a mine integrated energy system based on multi-task differential evolution as described in claim 6, characterized in that, In S42, the frequency domain difference mutation operator based on adaptive search weights includes two mutation operators: DE / current-to-pbest / 1 and DE / current-to-rand / 1. DE / current-to-pbest / 1 represents the difference between the current individual and the previous individual in the population. The differential evolution mutation operator learned by % of random elite individuals, where DE / current-to-rand / 1 is the differential evolution mutation operator learned by the current individual and random individuals in the population: Generate mutant individuals using DE / current-to-pbest / 1 The calculation formula is: （12） Generate mutant individuals using DE / current-to-rand / 1 The calculation formula is: （13） in, The first in the population Individual in the frequency domain, The top of the current population % of random individuals, , , These are three distinct random individuals in the current population. For the search weight function, the dimensions and same; No. Search weight function for each individual for: （14） in, It is the maximum number of generations the algorithm can evolve. This represents the current iteration number of the algorithm. For the weight vector, Include K The same weight vector , The dimension of each weight vector, for The Middle The weight values ​​of each component, It is a function determined by the frequency of the components. A function controlled by the number of iterations. and These are parameters used to control the steepness of the slope; parameter Updated using a Cauchy distribution obtained through random sampling, i.e. ; Position parameters The update method is as follows: （15） in, For the first Generational population The parameter values ​​for each individual, For the first The Lehmer mean of all individuals in the population over a generation.

8. The optimal scheduling method for a mine integrated energy system based on multi-task differential evolution according to claim 6, characterized in that, The specific steps of the S43 time-frequency conversion mechanism are as follows: S431. Transform from the time domain to the frequency domain, for time domain individuals of The discrete Fourier transforms are performed on the sub-variables to obtain the frequency domain representation of the time-domain individual; For the Group decision variables The discrete Fourier transform is expressed as follows: （16） in, For the first The original spectrum of the sub-variables, its dimension is , for The Middle Each variable can take a value. For the first in the spectrum One frequency component, For Fourier basis functions, The imaginary unit; S432. Transform from the frequency domain to the time domain; frequency domain individuals are... Spectrum Composition, first reconstruct the complete spectrum, for Perform inverse Fourier transform on each complete spectrum to convert it into a single spectrum. Each spectrum is reconstructed to the time domain to obtain the time-domain representation of the frequency-domain decision variables; The complete spectrum includes positive and negative frequency components, specifically represented as follows: （17） in, For conjugate operations, if It is an odd number, and the negative frequency component is composed of... The complex conjugates of the second to last components constitute; if If it is even, then take The second to second-to-last components are conjugated. The formula for the inverse Fourier transform is: （18） in, These are the basis functions of the inverse Fourier transform.

9. The optimal scheduling method for a mine integrated energy system based on multi-task differential evolution according to claim 6, characterized in that, The specific formula for the S44 boundary repair strategy is as follows: （19） in, For the first Individual, the first The values ​​of each decision component after boundary repair The upper boundary, This is the lower boundary.

Images