A new energy consumption promotion method based on a group incremental learning algorithm

Abstract

The application discloses a new energy consumption promotion method based on a group incremental learning algorithm, and the method comprises the following steps: (1) optimizing a multi-objective function of maximum wind and light consumption and minimum total operation cost in an electric energy scheduling model, and converting the optimized multi-objective function into a single objective function by using a linear weighting method; (2) establishing a constraint condition of the scheduling model, wherein the constraint condition comprises a system load balance constraint, a rotating load reserve constraint, a wind power and photovoltaic power generation operation constraint, and a hydroelectric generator and thermal power generator operation constraint; and (3) solving the optimized scheduling model based on an improved group incremental learning algorithm to obtain a new energy consumption scheme with the highest practical value. The method considers the dual objectives of maximum new energy consumption and minimum total operation cost, simultaneously considers operation constraint conditions of various power generators and system stability constraint conditions, and guarantees safe and reliable operation of the system under the optimal configuration scheme.

Description

Technical Field

[0001] This invention belongs to the field of renewable and clean energy development technology, and in particular relates to a method for improving the absorption of new energy sources based on a population incremental learning algorithm. Background Technology

[0002] The rapid development of the national economy has led to an increasing demand for electricity, while traditional thermal power generation emits greenhouse gases and pollutes the atmosphere. Renewable energy sources have gained widespread attention due to their cleanliness, pollution-free nature, and renewability, and countries worldwide are vigorously developing renewable energy power generation technologies. However, the randomness, intermittency, volatility, and anti-peak-shaving characteristics of renewable energy sources such as wind and solar power increase the peak-to-valley difference in the power grid, increasing the system's peak-shaving burden. To maintain the stability of the power system, wind and solar power often have to be curtailed, wasting a significant amount of natural resources.

[0003] On the other hand, regions rich in wind and solar power resources are often located in sparsely populated areas, far from load centers. Their local load capacity to absorb large-scale renewable energy is limited, and insufficient peak-shaving capacity has become a major obstacle to the large-scale development of renewable energy. Traditional dispatching models rely solely on conventional power sources for renewable energy absorption, which is insufficient to effectively solve the problem of optimized power system dispatching for renewable energy grid integration. New methods for optimizing power source allocation are urgently needed.

[0004] The complementary characteristics of various heterogeneous energy sources such as wind, solar, hydro, and thermal power can alleviate the randomness and volatility of wind and solar power generation to some extent, providing a new solution for improving the utilization rate of new energy sources. By utilizing the spatiotemporal complementarity among these heterogeneous energy sources, high-quality power output can be achieved, power supply reliability can be improved, and the utilization rate of renewable resources can be increased.

[0005] Wind and solar energy have excellent temporal complementarity. Generally, solar energy is stronger and wind energy is weaker in summer, on sunny days, and during the day, while wind energy is stronger and solar energy is weaker in winter, on cloudy days, and at night. Similarly, wind and hydropower also have good temporal complementarity. In my country, precipitation is concentrated in summer and less in winter, while wind energy is mainly concentrated in winter and spring, with less wind energy in summer.

[0006] Thermal power units have good peak-shaving capabilities and can maintain the stable operation of the power system by adjusting the output of the units. However, thermal power units have high start-up and shutdown costs and long start-up times. Hydropower units make up for the shortcomings of thermal power units and have the characteristics of fast start-up and shutdown and fast adjustment speed.

[0007] In conclusion, applying the spatiotemporal complementarity of heterogeneous energy sources such as wind, solar, hydro, and thermal power to joint operations can improve the reliability of power system operation, increase economic efficiency, reduce environmental pollution, and generate significant comprehensive benefits. Summary of the Invention

[0008] The purpose of this invention is to provide a new energy consumption enhancement method based on a population incremental learning algorithm. The aim is to minimize the total operating cost and maximize the total wind and solar energy consumption as optimization objectives. It takes into account system load balance constraints, spinning reserve capacity constraints, and related constraints of various heterogeneous energy generator sets such as wind, solar, hydro, and thermal. It utilizes the spatiotemporal complementary characteristics among wind, solar, hydro, and thermal to increase the power generation of renewable clean energy.

[0009] This invention is implemented as follows: a method for improving the absorption of new energy sources based on a population incremental learning algorithm, the method comprising:

[0010] (1) The multi-objective function of maximizing wind and solar energy consumption and minimizing total operating cost in the power dispatch model is optimized, and the optimized multi-objective function is transformed into a single-objective function by using the linear weighting method.

[0011] (2) Establish the constraints of the scheduling model, including system load balance constraints, spinning load reserve constraints, wind power and photovoltaic power generation operation constraints, and hydropower and thermal power unit operation constraints;

[0012] (3) Based on the improved group incremental learning algorithm, the optimal scheduling model is solved to obtain the new energy consumption scheme with the highest practical value.

[0013] Preferably, in step (1), the objective function that maximizes the wind and solar energy absorption is:

[0014]

[0015] Among them, P wp For total wind and solar energy absorption, P w (t), P p (t) represents the wind power output and photovoltaic power output at time t, respectively.

[0016] Preferably, in step (1), the objective function that minimizes the total operating cost is:

[0017]

[0018] Where C represents the total operating cost; C bi (t) represents the power generation cost per unit time of thermal power unit i; C hi (t), C w (t), C p (t) represents the maintenance cost per unit time for hydropower unit i, wind power unit, and photovoltaic power unit; N b N h These are the number of thermal power and hydropower units, respectively.

[0019] Preferably, in step (1), the single objective function is:

[0020]

[0021] Where C is the total operating cost, C min For total operating costs, P wp For total wind and solar energy absorption, P wpmax The maximum absorption capacity of total wind and solar power, P wpmin The minimum absorption capacity of the total wind and solar power is denoted by m1 and m2, which are weighting coefficients, and m1 + m2 = 1.

[0022] Preferably, in step (2), the system load balance constraint condition is:

[0023] P bi (t)+P hi (t)+P w (t)+P p (t)=P load (t) (4)

[0024] Among them, P load (t) represents the predicted load at time t;

[0025] The system spinning reserve capacity constraint is as follows:

[0026]

[0027] Among them, R t For the system's spinning reserve capacity, P bi,max P hi,max α, β, and γ represent the upper limits of output for thermal power and hydropower unit i, respectively, and the requirements of system load, wind power, and photovoltaic power generation prediction errors on spinning reserve, respectively.

[0028] The maximum output constraints for wind power and photovoltaic power generation are:

[0029] 0≤P w (t)≤min{p w (t),P w,max} (6)

[0030] 0≤P p (t)≤min{p p (t),P p,max} (7)

[0031] Where, p w (t), p p (t) represents the predicted maximum output of wind power and photovoltaic power at time t, respectively. w,max P p,max These are the upper limits for wind power and photovoltaic power generation, respectively;

[0032] The relevant constraints for hydropower are:

[0033] 0≤P hi (t)≤P hi,max (8)

[0034] P hi,max This is the upper limit of hydroelectric power output;

[0035] The relevant constraints for thermal power plants are:

[0036] P bi,min ≤P bi (t)≤P bi,max (9)

[0037] Among them, P bi,min P bi,max These are the upper and lower limits of the output of thermal power unit i, respectively.

[0038] Preferably, in step (3), the improvement of the population incremental learning algorithm includes the following:

[0039] A. Improvement of the probability matrix

[0040] The decision variables in each dimension are divided into M sub-intervals to define the probability matrix P. An intermediate solution is generated by the average value of the exploration solutions concentrated in the sub-intervals through the normal distribution function. The average value of the normal distribution function in a special interval is set as the weighted average value of the feasible solutions distributed in that interval.

[0041] B. Improvements to the probability matrix update

[0042] By recording the Pareto optimal solution for each individual and randomly selecting one from it when updating its probability distribution, an optimal balance between convergence and exploration can be obtained.

[0043] C. Improvements in search space refinement

[0044] The dimension of the feasible space is automatically refined in the proposed real-number encoded population incremental learning algorithm, which starts from a coarse discretization level and continues until an approximation of the Pareto optimal solution is obtained. After obtaining an approximate Pareto optimal solution at the current scale level, the algorithm refines its feasible search space by strengthening only those intervals in the subintervals of the Pareto optimal solution distribution until the stopping condition is met.

[0045] D. Improvements in fitness calculation

[0046] By introducing a metric, we can quantitatively assess and measure the number of improvement targets versus the amount of improvement for a specific target.

[0047] Preferably, the average value μ ik The improved algorithm is as follows:

[0048]

[0049] Where, N ik x is the number of previously explored feasible solutions distributed within a specified interval. ij Let be the j-th feasible solution within the interval, and fit() be the fitness function; for the minimum problem, the fitness function can be the reciprocal of the objective function; for the maximum problem, the fitness function is the objective function.

[0050] The improved algorithm for the probability matrix is ​​as follows:

[0051] LR ij =LR0exp(-(jr) 2 ) / (b i -a i )

[0052] p ij (t+1)=LR ij +(1-LR i )p ij (t)

[0053]

[0054] Where LR0 is the initial learning rate, and r is the number of variables. The interval index, where j is the interval index of the individual, b i and a i These are the upper and lower limits of the interval, p. ij (t) is the latest generation probability value, p ij (t+1) is the probability value of the next generation;

[0055] The improved algorithm for fitness calculation is as follows:

[0056] fit(x m ) = fit nor (x m )+Δ(x m (14)

[0057] Among them, fit nor (x m ) is the regularized x m Adaptability.

[0058] Compared with the shortcomings and deficiencies of existing technologies, the present invention has the following beneficial effects:

[0059] (1) The method of the present invention not only considers the dual objectives of maximizing the consumption of new energy and minimizing the total operating cost, but also considers the operating constraints of various generator sets and the system stability constraints, ensuring the safe and reliable operation of the system under the optimal configuration scheme;

[0060] (2) The method of the present invention is solved by the population incremental learning algorithm, which avoids getting stuck in local optima. In addition, based on the characteristics of bi-objective optimization, the population incremental learning algorithm is improved to make it applicable to solving multi-objective optimization problems. Attached Figure Description

[0061] Figure 1 This is a schematic diagram illustrating the principle of the population incremental learning algorithm for generating probability matrices.

[0062] Figure 2 These are typical daily load and wind and solar power output prediction results in the embodiments of the present invention;

[0063] Figure 3 This is the result of combined operation of wind, solar, water, and thermal power when all wind and solar energy are absorbed in the embodiments of the present invention;

[0064] Figure 4 This embodiment of the invention only considers the combined operation results of wind, solar, hydro, and thermal power when the total operating cost is minimized;

[0065] Figure 5 This is the result of the combined operation of wind, solar, hydro, and thermal power when considering maximizing the total wind and solar absorption capacity and minimizing the total operating cost in the embodiments of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0067] I. Implementation Plan

[0068] (I) Power Optimization Dispatch Model

[0069] To improve the utilization rate of renewable energy while minimizing power generation costs, the objective function of the optimized scheduling model in this invention is:

[0070] (1) The largest wind and solar energy absorption capacity

[0071] Wind and solar power absorption capacity refers to the total output of wind and solar power:

[0072]

[0073] Among them, P wp For total wind and solar energy absorption, P w (t), Pp (t) represents the wind power output and photovoltaic power output at time t, respectively.

[0074] (2) Minimum total operating cost

[0075] Total operating costs include the costs of generating electricity from thermal power units, as well as the costs of maintaining hydropower, wind power, and photovoltaic power units.

[0076]

[0077] Where C represents the total operating cost; C bi (t) represents the power generation cost per unit time of thermal power unit i; C hi (t), C w (t), C p (t) represents the maintenance cost per unit time for hydropower unit i, wind power unit, and photovoltaic power unit; N b N h These are the number of thermal power and hydropower units, respectively.

[0078] A linear weighted method is used to transform the multi-objective function into a single-objective function. During the transformation process, to ensure that each objective function has equal preference, a min-max normalization method is employed, resulting in the normalized single-objective function:

[0079]

[0080] Where m1 and m2 are weighting coefficients, and m1+m2=1.

[0081] (3) Constraints

[0082] The constraints of the optimized scheduling model in this paper include system load balance constraints, spinning load reserve constraints, wind power and photovoltaic power generation operation constraints, and hydropower and thermal power unit operation constraints.

[0083] 1) System load balance constraints

[0084] For the power system to operate smoothly and reliably, the first requirement is to ensure system load balance:

[0085] P bi (t)+P hi (t)+P w (t)+P p (t)=P load (t) (4)

[0086] Where P load (t) represents the predicted load at time t.

[0087] 2) System spinning reserve capacity constraints

[0088] Since wind and solar power output cannot be accurately predicted, a spin-off reserve capacity constraint needs to be added to ensure stable system operation.

[0089]

[0090] Among them, R t For the system's spinning reserve capacity, P bi,max P hi,max α, β, and γ represent the upper limits of output for thermal power and hydropower unit i, respectively, and the requirements of system load, wind power, and photovoltaic power generation prediction errors on spinning reserve, respectively.

[0091] 3) Maximum output constraints for wind and solar power generation

[0092] The operation of wind and solar power generators also needs to meet the upper and lower limits of the unit's output:

[0093] 0≤P w (t)≤min{p w (t),P w,max} (6)

[0094] 0≤P p (t)≤min{p p (t),P p,max} (7)

[0095] Where, p w (t), p p (t) represents the predicted maximum output of wind power and photovoltaic power at time t, respectively. w,max P p,max These are the upper limits for wind power and photovoltaic power generation, respectively.

[0096] 4) Hydropower-related constraints

[0097] The operation of hydroelectric generator units must meet the upper and lower limits of the unit's output:

[0098] 0≤P hi (t)≤P hi,max (8)

[0099] P hi,max This is the upper limit of hydropower output.

[0100] 5) Constraints related to thermal power

[0101] The operation of thermal power generating units must meet the upper and lower limits of the unit's output:

[0102] P bi,min ≤P bi (t)≤P bi,max (9)

[0103] Among them, P bi,min P bi,max These are the upper and lower limits of the output of thermal power unit i, respectively.

[0104] (II) Solving the Multi-Objective Optimization Scheduling Model Based on Population Incremental Learning Algorithm

[0105] The traditional population incremental learning algorithm (Reference: Baluja S, Caruana R. Removing the genetics from the standard genetic algorithm [M] / / Machine Learning Proceedings 1995. Morgan Kaufmann, 1995: 38-46.) is a binary-coded metaheuristic algorithm.

[0106] To extract global statistics from previous searches, a real-valued probability vector is introduced into the binary-coded population incremental learning algorithm. The length of this vector is the same as the encoded chromosome length of the feasible solution in the optimal problem. Each element of the vector corresponds to the probability that a chromosome position is encoded as 1, and this probability vector is used to generate a new population in each generation. After evaluating the objective function of the new population, this probability vector is updated only with the best individual in the current population, moving the vector closer to the chromosome of this new individual. Initially, all values ​​of the probability vector are set to 0.5, and as the search progresses, the probability vector is expected to gradually shift to correspond to the solution with the highest fitness value.

[0107] To make the swarm incremental learning algorithm applicable to multi-objective optimization problems, the traditional swarm incremental learning algorithm needs to be improved.

[0108] (1) Probability matrix

[0109] For the proposed real-number encoding, the real-valued probability vector in the binary-coded population incremental learning algorithm is represented by a probability matrix P. To define the probability matrix P, the decision variables in each dimension are divided into M sub-intervals. For example, the i-th variable x... i The j-th interval is defined as u ij =[a i +(j-1)Δ,a i +jΔ](Δ=(b i -a i ( / M). The value of probability matrix P. ij x represents i In the interval u ijThe probability matrix P is used to generate the entire population, where all values ​​are initially set to 1 / M to ensure that initial values ​​uniformly distributed in the feasible parameter space are generated from this probability matrix. i In the interval u ij The probability within p ij Proportional. To generate intermediate solutions that focus on the average of solutions explored in sub-intervals, a normal distribution function is proposed and used instead of the existing uniform distribution function. The implementation of this generation mechanism is as follows: Figure 1 As shown, x is a vector of decision variables; a i and b i These are the i-th variable x i The lower and upper bounds; It is a normal distribution function with a mean μ ik and standard deviation σ ik .

[0110] For population incremental learning algorithms with real-number encoding, the feasible solutions within a given interval are infinite, and the fitness values ​​of different feasible solutions may differ significantly even if they are distributed within the same interval. However, solutions with higher fitness values ​​are more likely to be the optimal point. In this regard, the average of the normal distribution function within a specific interval is set as the weighted average of the feasible solutions distributed within that interval; in other words, the average value μ... ik The expression is as follows:

[0111]

[0112] Where, N ik x is the number of previously explored feasible solutions distributed within a specified interval. ij It is the j-th feasible solution within the interval, and fit() is the fitness function. For the minimum value problem, the fitness function can be the reciprocal of the objective function; for the maximum value problem, the fitness function is the objective function.

[0113] (2) Probability matrix update

[0114] As mentioned earlier, in order to extract random information from the objective function to guide the search to a better region, the probability matrix P uses the population optimum x from the scalar population incremental learning algorithm. best Update. However, if the objective functions are treated equally, the optimal solution to a multi-objective optimization problem is not unique, which can lead to problems for population incremental learning algorithms. To achieve a good balance between convergence and exploration, each individual memorizes its "Pareto optimal solution" and randomly selects one when updating its probability distribution. Furthermore, the probability matrix is ​​updated at the end of each generation by the following equation:

[0115] LR ij=LR0exp(-(jr) 2 ) / (b i -a i )

[0116] p ij (t+1)=LR ij +(1-LR i )p ij (t)

[0117]

[0118] Where LR0 is the initial learning rate, and r is the number of variables. Range index.

[0119] (3) Refinement of the search space

[0120] In implementing the proposed real-number encoded population incremental learning algorithm, small intervals should be used to produce high-precision final results. For high-dimensional optimization problems, small intervals lead to extremely large probability matrices, reducing the algorithm's convergence performance. However, if the number of intervals in each dimension is too small, even if the convergence performance is improved, the sampling of feasible solutions from the corresponding probability matrices will be coarse, resulting in low precision of the final solution. To achieve a good balance between convergence speed and solution quality, a multi-scale vector population incremental learning algorithm is proposed. More specifically, the dimension of the feasible space is automatically refined in the proposed real-number encoded population incremental learning algorithm. The algorithm starts from a coarse discretization level and continues until an approximation of the Pareto optimal solution is obtained. After obtaining an approximate Pareto optimal solution at the current scale level, the algorithm refines its feasible search space by strengthening only those intervals in the sub-intervals of the "Pareto optimal solution" distribution until the stopping condition is met.

[0121] (4) Fitness calculation

[0122] In existing vector evolution algorithms, the mainstream approach is to assign fitness to individuals in a multi-objective optimization problem. However, this method may only qualitatively determine the advantage relationships, but cannot quantitatively measure the number of objectives that can be improved, and the improvement of a specific objective may also be unquantifiable. To address this issue, a metric is first introduced to quantitatively evaluate the improvements in both aspects. For example, for minimizing multi-objective optimization, the proposed evaluation metric x... m Relative to solution x n The improvement is given by the following formula.

[0123]

[0124] Then, the solution x is given. m The total improvement relative to other individuals in the current group is

[0125]

[0126] Solution x m The fitness value is ultimately determined by the following formula.

[0127] fit(x m ) = fit nor (x m )+Δ(x m (14)

[0128] Among them, fit nor (x m ) is the regularized x m Adaptability.

[0129] II. Experimental Verification

[0130] To demonstrate the effectiveness of the method of this invention, this invention uses a power system in a certain region as an example to perform optimization analysis and calculation using the aforementioned model and algorithm. This region has 10 thermal power units and 3 hydropower units, which are hydropower plants with daily regulation characteristics. Unit parameters are shown in Table 1, and hydropower station parameters are shown in Table 2. The maintenance costs for hydropower, wind power, and photovoltaic generator units are 26.7 yuan / MWh, 29.6 yuan / MWh, and 19.6 yuan / MWh, respectively. A typical daily load and wind / solar output prediction curve are shown in Table 2. Figure 2 .

[0131] Table 1 Parameters of Hydropower Units

[0132]

[0133] Table 2 Hydropower Station Parameters

[0134]

[0135] Based on the given unit parameters and Figure 2 Based on the predicted load and wind and solar power output values, and combined with the multi-objective power system optimization scheduling model proposed in this paper based on the complementary characteristics of multiple heterogeneous energy sources, the simulated annealing method is applied to optimize and analyze the following three different operating conditions:

[0136] 1) When all wind and solar energy are absorbed, wind, solar, water, and thermal power will operate in combination;

[0137] 2) Only consider the combined operation of wind, solar, hydro, and thermal power when the total operating cost is minimized;

[0138] 3) Consider the large total wind and solar energy consumption and the total operating cost of combined wind, solar, hydro, and thermal power operation.

[0139] Simulation results are as follows Figures 3-5As shown in Tables 3 and 4, the calculation results show that when only the lowest cost is considered, thermal power output increases while wind power output decreases significantly, resulting in severe wind curtailment. When both the goals of maximizing renewable energy consumption and minimizing total operating costs are considered simultaneously, costs increase by approximately 12.2%, but the wind curtailment rate decreases significantly, reaching only 0.19% of the rate when only the lowest cost is considered, while thermal power output decreases by approximately 27.4%. Figures 3-5 The calculation results show that the combined operation of multiple heterogeneous energy sources can reduce generalized load fluctuations to a certain extent, increase the total wind and solar energy consumption, and reduce the output of thermal power units.

[0140] Table 3 Costs and Thermal Power Unit Outputs under Different Operating Conditions

[0141]

[0142] Table 4. Wind and Solar Curtailment Amounts under Different Operating Conditions

[0143]

[0144]

[0145] III. Conclusion

[0146] This invention develops a method for improving renewable energy consumption with the dual objectives of maximizing renewable energy absorption and minimizing total operating costs through a population incremental learning algorithm. Figure 5 The comparison results under the three operating conditions shown in Tables 3 and 4 demonstrate that the method proposed in this invention can simultaneously maximize the absorption of new energy power and reduce operating costs. The method proposed in this invention has the following advantages:

[0147] 1. The method of the present invention not only considers the dual objectives of maximizing the consumption of new energy and minimizing the total operating cost, but also considers the operating constraints of various types of generator sets and the system stability constraints, ensuring the safe and reliable operation of the system under the optimal configuration scheme;

[0148] 2. The method of this invention uses a population incremental learning algorithm to solve the problem, which avoids getting stuck in local optima. In addition, based on the characteristics of bi-objective optimization, the population incremental learning algorithm is improved to make it applicable to solving multi-objective optimization problems.

[0149] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for improving the absorption and utilization of new energy sources based on a population incremental learning algorithm, characterized in that, The method includes: (1) The multi-objective function of maximizing wind and solar energy consumption and minimizing total operating cost in the power dispatch model is optimized, and the multi-objective function is transformed into a single objective function by using the linear weighting method; (2) Establish the constraints of the scheduling model, including system load balance constraints, spinning load reserve constraints, wind power and photovoltaic power generation operation constraints, and hydropower and thermal power unit operation constraints; (3) The optimized scheduling model is solved based on the improved group incremental learning algorithm to obtain the new energy consumption scheme; In step (3), the improvement to the population incremental learning algorithm includes the following: A. Improvement of the probability matrix The decision variables in each dimension are divided into M sub-intervals to define the probability matrix P. An intermediate solution is generated by the average value of the exploration solutions concentrated in the sub-intervals through the normal distribution function. The average value of the normal distribution function in a special interval is set as the weighted average value of the feasible solutions distributed in that interval. B. Improvements to the probability matrix update By recording the Pareto optimal solution for each individual and randomly selecting one from it when updating its probability distribution, an optimal balance between convergence and exploration can be obtained. C. Improvements in search space refinement The dimension of the feasible space is automatically refined in the proposed real-number encoded population incremental learning algorithm. The algorithm starts from a coarse discretization level and continues until an approximation of the Pareto optimal solution is obtained. After obtaining an approximate Pareto optimal solution at the current scale level, the algorithm refines its feasible search space by strengthening only those intervals in the subintervals of the Pareto optimal solution distribution until the stopping condition is met. D. Improvements in fitness calculation By introducing a metric, we can quantitatively assess the number of improvement targets and the amount of improvement for a specific target.

2. The method as described in claim 1, characterized in that, In step (1), the objective function that maximizes the wind and solar energy absorption is: （1） in, This refers to the total amount of wind and solar energy absorbed. , They are respectively Wind power output and solar power output at all times.

3. The method as described in claim 1, characterized in that, In step (1), the objective function that minimizes the total operating cost is: （2） in, Total operating cost; For thermal power units The cost of generating electricity per unit of time; , , For hydroelectric generator units Maintenance costs per unit time for wind power and photovoltaic generator sets; , These represent the number of thermal power and hydropower units, respectively.

4. The method as described in claim 1, characterized in that, In step (1), the single objective function is: （3） Where C represents the total operating cost. For total wind and solar energy absorption, P wpmax P is the maximum absorption capacity of total wind and solar power. wpmin This is the minimum amount of total wind and solar energy that can be absorbed. and These are the weighting coefficients, and .

5. The method as described in claim 1, characterized in that, In step (2), the system load balancing constraint is: （4） in, P w (t) represents the wind power output at time t; P p (t) represents the photovoltaic output at time t; for Forecasted load at any given time; The system spinning reserve capacity constraint is as follows: （5） in, For system spin-off reserve capacity, , They are thermal power and hydropower units. The upper limit of output, N b N represents the number of thermal power units. h Indicates the number of hydroelectric generating units. , , These are the requirements for spinning reserve based on system load, wind power, and photovoltaic power generation forecast errors, respectively. The maximum output constraints for wind power and photovoltaic power generation are: （6） （7） in, , They are respectively Predicted maximum output of wind and solar power at any given time. , These are the upper limits for wind power and photovoltaic power generation, respectively; The relevant constraints for hydropower are: （8） This represents the upper limit of hydroelectric power output. The relevant constraints for thermal power plants are: （9） in, , thermal power units The upper and lower limits of output.

6. The method as described in claim 1, characterized in that, The average value μ ik The improved algorithm is as follows: （10） Where, N ik x is the number of previously explored feasible solutions distributed within a specified interval. ij Let be the j-th feasible solution within the interval, and fit() be the fitness function; for the minimum problem, the fitness function is the reciprocal of the objective function; for the maximum problem, the fitness function is the objective function. The improved algorithm for the probability matrix is ​​as follows: （11） Where LR0 is the initial learning rate, and r is the number of variables containing the initial learning rate. The interval index, where j is the interval index of the individual, b i and a i These are the upper and lower limits of the interval, p. ij (t) is the latest generation probability value, p ij (t+1) is the probability value of the next generation; The improved algorithm for fitness calculation is as follows: （14） in, It is the regularized x m Adaptability.

Images