Large power grid reactive power optimization method and device, storage medium and computer equipment

Abstract

According to the large power grid reactive power optimization method and device, the storage medium and the computer equipment provided by the invention, the advantages of the two algorithms are fully exerted through the hybrid chaos quantum particle swarm optimization algorithm and the dimension-by-dimension convex space search algorithm. According to the chaotic quantum particle swarm algorithm, the global search capability and the capability of jumping out of local optimum of a particle swarm are enhanced by utilizing the characteristics of quantum behaviors and chaotic mapping, and the problem of premature convergence of a traditional heuristic intelligent algorithm is avoided. And according to the dimension-by-dimension convex space search algorithm, fine search is carried out on each excellent particle in different dimensions, a local optimal solution is determined, and the search precision and efficiency are further improved. According to the design of the hybrid algorithm, special optimization is carried out aiming at the characteristics of a reactive power optimization problem model, such as variable property difference, constraint complexity and the like, and the technical defects of poor optimization effect and optimization efficiency of an optimization solution algorithm in the prior art are effectively overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power systems and their automation, and particularly relates to a large power grid reactive power optimization method and device, a storage medium and computer equipment. BACKGROUND

[0002] Large power grid reactive power optimization is a classic mixed integer nonlinear optimization problem, which aims to minimize the active power loss of the system and maintain the voltage level within the qualified range by adjusting the control variables under the premise of meeting the safe operation constraints of the power grid.

[0003] The existing optimization solution algorithms are mainly divided into three categories: traditional mathematical programming methods, heuristic\intelligent algorithms, and hybrid algorithms. Among them, traditional mathematical programming methods such as primal-dual interior point method and nonlinear programming method have high requirements for the convexity and continuity of the objective function and the constraint conditions, and it is difficult to directly process discrete variables (usually processed by rounding after continuous processing, resulting in suboptimal or even infeasible results); heuristic\intelligent algorithms such as genetic algorithm, particle swarm algorithm, and differential evolution algorithm do not rely on gradient information and can handle discrete variables and non-convex problems, but generally have problems such as premature convergence (falling into local optimum), slow convergence speed in the later stage, parameter sensitivity, and sharp decline in search efficiency in high-dimensional space; hybrid algorithms, such as particle swarm algorithm based on interior point method, combine interior point method and particle swarm algorithm, but only perform simple alternation between the two, without deep fusion and complementary advantages, and lack of special design for the characteristics of the reactive power optimization problem model (such as variable property difference and constraint complexity).

[0004] Therefore, it is urgent to develop a new solution algorithm that can overcome the above-mentioned defects, has strong global exploration ability and fine local development ability, and can efficiently handle mixed variables and complex constraints. SUMMARY

[0005] The purpose of the present application is to at least solve one of the above technical defects, in particular the technical defect that the optimization effect and optimization efficiency of the existing large power grid reactive power optimization solution algorithm are poor.

[0006] The present application provides a large power grid reactive power optimization method, which comprises:

[0007] Obtaining the power grid topology, line parameters, load data, generator operating limits, and voltage safety limits of the target area, and setting optimization algorithm parameters;

[0008] According to the power grid topology, the line parameters, the load data, the generator operating limits, and the voltage safety limits, constructing an objective function for minimizing the total network loss and corresponding constraint conditions, the constraint conditions including upper and lower limit constraints corresponding to the decision variables;

[0009] randomly initialize a particle population based on the upper and lower limit constraints corresponding to the decision variables and the optimization algorithm parameters, each particle in the particle population containing continuous encoding of all continuous variables and discrete variables, and one particle position representing one reactive power control scheme;

[0010] based on the hybrid chaotic quantum particle swarm and the dimension-by-dimension convex space search algorithm, optimize and solve the particle position of each particle in the particle population to obtain a global optimal position that minimizes the total network loss, and take the global optimal position as the optimal control scheme during power grid dispatching.

[0011] Optionally, based on the hybrid chaotic quantum particle swarm and the dimension-by-dimension convex space search algorithm, optimizing and solving the particle position of each particle in the particle population to obtain a global optimal position that minimizes the total network loss, comprises:

[0012] update the historical optimal position of each particle in the particle population and the global optimal position of the particle population based on the constraint violation degree;

[0013] based on the updated historical optimal position of each particle and the global optimal position of the particle population, and based on the quantum behavior and the chaotic mapping particle swarm optimization mechanism, perform secondary update on the particle position of all particles in the particle population;

[0014] based on the dimension-by-dimension convex space search algorithm and the particle position of each particle after the secondary update, determine the local optimal solution of excellent particles in the particle population in different dimensions, and then return to perform the feasibility rule based on the constraint violation degree to update the historical optimal position of each particle in the particle population and the global optimal position of the particle population and the subsequent steps until the optimization algorithm parameters are met, thereby obtaining the global optimal position that minimizes the total network loss.

[0015] Optionally, the feasibility rule based on the constraint violation degree updates the historical optimal position of each particle in the particle population and the global optimal position of the particle population, comprising:

[0016] for each particle in the particle population, convert the continuous encoding of the discrete variables in the particle into actual control instructions of the power grid;

[0017] perform power flow calculation according to the continuous encoding of the continuous variables in the particle and the actual control instructions to obtain a power flow calculation result;

[0018] determine the fitness and the constraint violation degree of the particle according to the power flow calculation result;

[0019] According to the fitness and the constraint violation degree of the particle, and based on a feasibility rule of the constraint violation degree, the historical optimal position of the particle and the global optimal position of the particle population are updated.

[0020] Optionally, the constraint condition comprises an equality constraint and an inequality constraint.

[0021] The fitness and the constraint violation degree of the particle are determined according to the power flow calculation result, comprising:

[0022] A power flow residual that is generated in a power flow calculation process and is not completely satisfied is acquired.

[0023] The constraint violation degree of the particle is calculated according to the power flow residual and a total exceeding amount of all inequality constraints.

[0024] The total active network loss of the whole network is calculated according to the power flow calculation result and the objective function.

[0025] The fitness of the particle is calculated based on the total active network loss, a first penalty term corresponding to the inequality constraint, and a second penalty term corresponding to the equality constraint.

[0026] Optionally, the historical optimal position of the particle and the global optimal position of the particle population are updated according to the fitness and the constraint violation degree of the particle, and based on a feasibility rule of the constraint violation degree, comprising:

[0027] The current position, the fitness and the constraint violation degree of the particle are compared with the historical optimal position, the fitness and the constraint violation degree of the historical optimal position of the particle, to obtain a first comparison result, and the historical optimal position of the particle is updated according to the first comparison result, based on a feasibility rule of the constraint violation degree.

[0028] The historical optimal positions of all particles are compared with each other to obtain a second comparison result, and the global optimal position of the particle population is updated according to the second comparison result, based on the feasibility rule of the constraint violation degree.

[0029] Optionally, the particle positions of all particles in the particle population are updated again according to the updated historical optimal positions of the particles and the global optimal position of the particle population, and based on a particle swarm optimization mechanism of quantum behavior and chaotic mapping, comprising:

[0030] A center point of the historical optimal positions of all particles is calculated according to the updated historical optimal positions of the particles.

[0031] For each particle, a random position between the updated historical optimal positions of the particles and the global optimal position of the updated particle population is generated.

[0032] generating a chaotic sequence using a Tent chaotic mapping formula, and determining a current shrinkage and expansion coefficient of each particle according to the chaotic sequence;

[0033] updating particle positions of all particles in the particle population based on the center point, the random position of each particle, and the current shrinkage and expansion coefficient.

[0034] Optionally, determining local optimal solutions of excellent particles in the particle population in different dimensions based on the dimension-by-dimension convex space search algorithm and the particle positions of the particles after the secondary update, comprises:

[0035] determining excellent particles in the particle population and one-dimensional convex intervals of each excellent particle in different dimensions according to fitness of each particle in the particle population;

[0036] searching in the one-dimensional convex intervals of each excellent particle in different dimensions using a golden section method to determine dimension values in the one-dimensional convex intervals that make the objective function optimal, and taking the dimension values of each excellent particle in the one-dimensional convex intervals as local optimal solutions of each excellent particle in different dimensions.

[0037] The application further provides a large power grid reactive power optimization device, comprising:

[0038] a data acquisition module configured to acquire power grid topology, line parameters, load data, generator operation limits, and voltage safety limits of a target region, and set optimization algorithm parameters;

[0039] a function construction module configured to construct an objective function and corresponding constraint conditions that make the total network loss minimum according to the power grid topology, the line parameters, the load data, the generator operation limits, and the voltage safety limits, wherein the constraint conditions comprise upper and lower limit constraints corresponding to decision variables;

[0040] an initialization module configured to randomly initialize a particle population based on the upper and lower limit constraints corresponding to the decision variables and the optimization algorithm parameters, wherein each particle in the particle population comprises continuous encoding of all continuous variables and discrete variables, and one particle position represents one reactive power control scheme;

[0041] an optimization solution module configured to optimize and solve particle positions of each particle in the particle population based on a hybrid chaotic quantum particle swarm and dimension-by-dimension convex space search algorithm, to obtain a global optimal position that makes the total network loss minimum, and take the global optimal position as an optimal control scheme during power grid dispatching.

[0042] The application further provides a computer-readable storage medium, wherein computer-readable instructions are stored in the computer-readable storage medium, and the computer-readable instructions, when executed by one or more processors, cause the one or more processors to perform the steps of the power grid reactive power optimization method according to any one of the above embodiments.

[0043] The application further provides a computer device, comprising: one or more processors, and a memory.

[0044] The memory stores computer-readable instructions, and the computer-readable instructions, when executed by the one or more processors, perform the steps of the power grid reactive power optimization method according to any one of the above embodiments.

[0045] As can be seen from the above technical solutions, the embodiments of the application have the following advantages:

[0046] The power grid reactive power optimization method, device, storage medium and computer device provided by the application can, after obtaining the power grid topology, line parameters, load data, generator operating limit value and voltage safety limit value of a target area and setting optimization algorithm parameters, construct a target function and corresponding constraint conditions for minimizing the total network loss according to the power grid topology, line parameters, load data, generator operating limit value and voltage safety limit value. Since the constraint conditions include upper and lower limit constraints corresponding to decision variables, the application can randomly initialize a particle population based on the upper and lower limit constraints corresponding to the decision variables and the optimization algorithm parameters. Each particle in the particle population contains continuous coding of all continuous variables and discrete variables, and a particle position represents a reactive power control scheme. Then, the application can optimize and solve the particle position of each particle in the particle population based on the hybrid chaotic quantum particle swarm algorithm and the dimension-by-dimension convex space search algorithm to obtain a global optimal position for minimizing the total network loss. The global optimal position can be used as an optimal control scheme for power grid dispatching. In the above process, the hybrid chaotic quantum particle swarm algorithm and the dimension-by-dimension convex space search algorithm fully exert the advantages of the two algorithms. The chaotic quantum particle swarm algorithm utilizes the characteristics of quantum behavior and chaotic mapping to enhance the global search ability and the ability to jump out of local optimum of the particle swarm, and avoids the premature convergence problem of traditional heuristic intelligent algorithms. The dimension-by-dimension convex space search algorithm performs fine search on different dimensions for each excellent particle to determine a local optimal solution, further improving the search accuracy and efficiency. The design of this hybrid algorithm is specially optimized for the characteristics of the reactive power optimization problem model, such as variable property differences and constraint complexity, effectively overcoming the technical defects of poor optimization effect and efficiency of the optimization solving algorithm in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the accompanying drawings in the following description only represent some embodiments of the present application, and for those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0048] Figure 1 A flowchart of a large power grid reactive power optimization method provided by an embodiment of the present application;

[0049] Figure 2 A flowchart of calculating a global optimal position provided by an embodiment of the present application;

[0050] Figure 3 A structural diagram of a large power grid reactive power optimization device provided by an embodiment of the present application;

[0051] Figure 4 An internal structural diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments only represent some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort fall within the scope of protection of the present application.

[0053] In one embodiment, as shown in Figure 1 , Figure 1 A flowchart of a large power grid reactive power optimization method provided by an embodiment of the present application; the present application provides a large power grid reactive power optimization method, which can include:

[0054] S110: Obtain the power grid topology, line parameters, load data, generator operating limit value and voltage safety limit value of the target area, and set the optimization algorithm parameters.

[0055] In this step, when performing reactive power optimization on the power grid dispatching of the target area, the power grid topology, line parameters, load data, generator operating limit value and voltage safety limit value of the target area can be obtained first, and the optimization algorithm parameters can be set, so that the large power grid reactive power optimization can be performed according to the obtained related data.

[0056] The power grid topology obtained by the application refers to the connection relationship and layout structure between elements in the target area power grid, which describes in detail the physical connection mode and spatial distribution of key elements such as power plants, substations, transmission lines and load points in the power grid. The line parameters include the electrical characteristic parameters of the transmission line such as resistance, reactance, conductance and susceptance, which are crucial for accurately calculating the power grid power flow and network loss. The load data reflects the power demand of the target area at each time, including the size and change law of active load and reactive load, and is one of the important factors to be considered in reactive power optimization. The generator operating limit specifies the upper and lower limit ranges of active power, reactive power, voltage and frequency parameters that must be followed during the operation of the generator to ensure the safe and stable operation of the generator. The voltage safety limit specifies the allowable fluctuation range of the voltage of each node in the power grid to prevent damage to power grid equipment and users caused by excessively high or low voltage.

[0057] Further, the application sets optimization algorithm parameters to control and adjust the performance and behavior of the algorithm in the subsequent optimization solving process, such as the size of the particle population, the number of iterations, the convergence accuracy, etc. The selection of these parameters will directly affect the convergence speed and solution quality of the optimization algorithm. By obtaining these basic data and setting the optimization algorithm parameters, the necessary data support and algorithm configuration are provided for subsequent construction of the objective function and constraints and optimization solving.

[0058] S120: According to the power grid topology, line parameters, load data, generator operating limit and voltage safety limit, an objective function for minimizing the total network loss and corresponding constraints are constructed, and the constraints include the upper and lower limit constraints of the decision variables.

[0059] In this step, after obtaining the power grid topology, line parameters, load data, generator operating limit and voltage safety limit of the target area through S110, the application can construct an objective function for minimizing the total network loss and corresponding constraints according to the power grid topology, line parameters, load data, generator operating limit and voltage safety limit. In this way, the objective function can be optimized and solved according to the objective function and constraints, and the optimal reactive power control scheme can be determined.

[0060] In a specific implementation, the large power grid reactive power optimization problem aims to arrange local compensation and balance of reactive power and avoid long-distance transmission of reactive power, so as to achieve the purpose of reducing the total network loss. Therefore, the objective function of the large power grid reactive power optimization problem in the application can be:

[0061]

[0062] wherein, is the total network loss of the power grid, and These refer to the active power generated by generators (including pumped storage and energy storage in power generation mode) and the active power absorbed by loads (including pumped storage and energy storage in charging mode).

[0063] The constraints in this application include power flow equations and upper and lower bound constraints on variables.

[0064] The active and reactive power flow equations of the power grid are as follows:

[0065]

[0066]

[0067] in, and These are the active power and reactive power of the balancing node, respectively. and These are the active power and reactive power of the power generation node, respectively. and These are the active power and reactive power of the load node, respectively. and These are the charging power and discharging power of the energy storage device, respectively. This refers to reactive power compensation. and This refers to the node voltage amplitude. and The admittance value of the Jacobian matrix of the power grid; This refers to the voltage phase angle; This represents the number of nodes in the power grid.

[0068] The variable upper and lower bound constraints in this application include those mentioned above. and , and , and , and , , and Upper and lower limit constraints.

[0069] The decision variables include generator reactive power. Node reactive power compensation And the transformer turns ratio T. Among these, the generator reactive power... and node reactive power compensation The transformer turns ratio T is not directly included in the reactive power flow equation, but is indirectly included in the power flow equation. The Jacobian matrix admittance value is directly included in the reactive power flow equation. and middle.

[0070] S130: Based on the upper and lower bound constraints corresponding to the decision variables and the optimization algorithm parameters, a particle swarm is randomly initialized. Each particle in the particle swarm contains the continuous codes of all continuous and discrete variables, and one particle position represents a reactive power control scheme.

[0071] In this step, after constructing the objective function and corresponding constraints that minimize the network loss of the entire network through S120, the constraints include the upper and lower bound constraints corresponding to the decision variables. Therefore, this application can randomly initialize the particle swarm according to the upper and lower bound constraints corresponding to the decision variables and the optimization algorithm parameters. Each particle in the particle swarm contains the continuous encoding of all continuous and discrete variables, and one particle position represents a reactive power control scheme. In this way, the particle position of each particle in the particle swarm can be optimized and solved based on the hybrid chaotic quantum particle swarm and the dimensional convex space search algorithm to obtain the optimal control scheme.

[0072] Specifically, when randomly initializing the particle swarm, this application allocates a corresponding encoding space to each decision variable based on its type and value range. For continuous variables, real-number encoding is used to ensure continuous variation within their value range, thus fully exploring the possibilities in the solution space. For discrete variables, integer or binary encoding is used, based on the specific discrete values ​​of the variable, to ensure the accuracy of the discrete variable's values. During the encoding process, the upper and lower bound constraints corresponding to the decision variables are strictly followed to ensure that the encoding of each particle is within a reasonable range. Furthermore, this application can combine the particle swarm size set in the optimization algorithm parameters to generate a corresponding number of particles. Each particle contains complete continuous encodings for all continuous and discrete variables, and each particle position represents a complete reactive power control scheme, thereby providing an initial particle swarm for subsequent optimization solutions.

[0073] For example, the particles generated by the method described above in this application can be represented as [ , The vector [T1, T2, ...], where, It is the generator's reactive power (continuous). T1, T2, ... are the reactive power compensation at the nodes (continuously encoded, but actually discrete), and T1, T2, ... are the transformer turns ratios (continuously encoded, but actually discrete).

[0074] S140: Based on the hybrid chaotic quantum particle swarm optimization and the dimensional convex space search algorithm, the particle position of each particle in the particle swarm is optimized to obtain the global optimal position that minimizes the network loss. The global optimal position is then used as the optimal control scheme for power grid dispatch.

[0075] In this step, after the particle population is randomly initialized based on the upper and lower limit constraints corresponding to the decision variables and the optimization algorithm parameters through S130, the application can optimize and solve the particle position of each particle in the particle population based on the hybrid chaotic quantum particle swarm and dimension-by-dimension convex space search algorithm, and then obtain the global optimal position that minimizes the total network loss. The application can take the global optimal position as the optimal control scheme for power grid dispatching.

[0076] It can be understood that the chaotic quantum particle swarm algorithm and the dimension-by-dimension convex space search algorithm exhibit good synergistic effects in the optimization and solving process. The chaotic quantum particle swarm algorithm can quickly perform global exploration in the solution space by virtue of the randomness of quantum behavior and the ergodicity of chaotic mapping, effectively avoiding the dilemma of falling into a local optimal solution. It adjusts the position and speed of the particles constantly, so that the particle swarm gradually approaches the global optimal solution, providing a good foundation for subsequent fine search.

[0077] The dimension-by-dimension convex space search algorithm, on the other hand, conducts in-depth and fine search on each excellent particle in different dimensions based on this. It utilizes the properties of convex space to decompose complex optimization problems into multiple one-dimensional sub-problems and solves them one by one to determine the local optimal solution. This dimension-by-dimension search approach greatly improves the accuracy and efficiency of the search, enabling further excavation of better solutions in the vicinity of the approximate optimal region found by the chaotic quantum particle swarm algorithm.

[0078] Through the organic combination of the two algorithms, the hybrid algorithm proposed by the application can fully leverage the advantages of each algorithm and compensate for the shortcomings of the other. In dealing with large power grid reactive power optimization problems, the hybrid algorithm can effectively overcome the problems of poor optimization effect and optimization efficiency in existing optimization solving algorithms, providing more accurate and efficient optimal control schemes for power grid dispatching.

[0079] For example, the application can apply the large power grid reactive power optimization method based on the hybrid chaotic quantum particle swarm and dimension-by-dimension convex space search to a provincial power grid to solve traditional reactive power optimization problems. The load capacity of this provincial power grid exceeds 160,000 MW, and the number of nodes at different voltage levels exceeds 20,000, so the optimization scale is large and the solution is difficult. The installed capacity of various power sources is shown in the following table:

[0080] Table 1 Installed capacity of various power sources

[0081]

[0082] After applying the reactive power optimization method of the application, the comparison results shown in Table 2 can be obtained:

[0083] Table 2 Comparison of results of the reactive power optimization method of the application and other optimization algorithms

[0084]

[0085] As shown in Table 2, compared with optimization algorithms such as genetic algorithm, classical particle swarm optimization, reinforcement learning algorithm, and interior-point method, the reactive power optimization method proposed in this application achieves the lowest network loss power, indicating that it is closest to the global optimum and has the best optimization effect. This also proves that the improvement proposed in this application to the classical particle swarm optimization algorithm is effective. Furthermore, the optimization method proposed in this application requires the least time for reactive power optimization among these methods, exhibiting the highest optimization efficiency. Therefore, it is more suitable for reactive power optimization problems in large power grids.

[0086] In one embodiment, such as Figure 2 As shown, Figure 2 This is a flowchart illustrating the calculation of the globally optimal position provided in an embodiment of this application; in S140, based on a hybrid chaotic quantum particle swarm optimization and dimensional convex space search algorithm, the particle position of each particle in the particle swarm is optimized to obtain the globally optimal position that minimizes the network loss. This may include:

[0087] S141: Based on the feasibility rules of constraint violation, update the historical optimal position of each particle in the particle population and the global optimal position of the particle population.

[0088] S142: Based on the updated historical optimal position of each particle and the global optimal position of the particle swarm, and based on the particle swarm optimization mechanism of quantum behavior and chaotic mapping, the particle positions of all particles in the particle swarm are updated a second time.

[0089] S143: Based on the dimensional convex space search algorithm and the particle positions of each particle after the second update, after determining the local optimal solutions of the excellent particles in the particle population in different dimensions, return to execute the feasibility rules based on the constraint violation degree, update the historical optimal position of each particle in the particle population and the global optimal position of the particle population and its subsequent steps, until the optimization algorithm parameters are satisfied, and obtain the global optimal position that minimizes the network loss of the entire network.

[0090] In this embodiment, when calculating the globally optimal position that minimizes the network loss, the historical optimal position of each particle in the particle swarm and the globally optimal position of the particle swarm can first be updated based on the feasibility rules of constraint violation. This step ensures that the movement of particles in the solution space always satisfies the given constraints, avoids invalid searches, and improves optimization efficiency.

[0091] Subsequently, the application can use the particle swarm optimization mechanism of quantum behavior and chaotic mapping to update the particle positions of all particles in the particle swarm according to the updated historical optimal positions of each particle and the global optimal position of the particle swarm. The randomness of quantum behavior and the ergodicity of chaotic mapping enable the particles to explore the solution space more extensively and deeply, which helps to find the global optimal solution.

[0092] Then, the application can determine the local optimal solution of the excellent particle in the particle swarm in different dimensions based on the dimension-by-dimension convex space search algorithm and the particle positions of each particle after the secondary update. The dimension-by-dimension convex space search algorithm decomposes the complex optimization problem into multiple one-dimensional sub-problems and solves them one by one, thereby finding the local optimal solution in each dimension and further improving the accuracy and efficiency of the search.

[0093] After determining the local optimal solution, the algorithm returns to perform step S141, i.e., updating the historical optimal position of each particle in the particle swarm and the global optimal position of the particle swarm based on the feasibility rule of constraint violation degree again. This process is repeated until the termination condition set in the optimization algorithm parameters is met, such as reaching the maximum number of iterations or the change of the global optimal position being less than a certain threshold.

[0094] Finally, when the termination condition is met, the algorithm outputs the global optimal position that minimizes the network loss, which can be used as the optimal control scheme for power grid dispatching. The optimal control scheme can include the optimal reactive power output setting value of each generator node, ensuring that the generator meets its own operating constraints while providing appropriate reactive power support for the power grid; the optimal switching amount of the reactive power compensation device near each load node, which accurately adjusts the reactive power distribution in the load area and improves the voltage quality; and the optimal adjustment position of each transformer tap, which optimizes the power flow distribution of the power grid by changing the transformer ratio to reduce line transmission loss. These specific control parameters collectively constitute a complete and efficient reactive power control strategy that can dynamically adjust the distribution and transmission of reactive power according to the real-time operating state of the power grid, achieving the goal of minimizing the network loss and effectively improving the economy and stability of large power grid operation.

[0095] Through these steps, the large power grid reactive power optimization method proposed by the application can efficiently and accurately find the optimal reactive power control scheme in a large-scale power grid, providing strong support for the safe, stable, and economic operation of the power grid.

[0096] In one embodiment, the feasibility rule based on the constraint violation degree in S141 for updating the historical optimal position of each particle in the particle swarm and the global optimal position of the particle swarm can include:

[0097] S1411: For each particle in the particle population, convert the continuous encoding of the discrete variable in the particle into an actual control instruction of the power grid.

[0098] S1412: Perform a power flow calculation according to the continuous encoding of the continuous variable in the particle and the actual control instruction, to obtain a power flow calculation result.

[0099] S1413: Determine the fitness and constraint violation degree of the particle according to the power flow calculation result.

[0100] S1414: Update the historical optimal position of the particle and the global optimal position of the particle population according to the fitness and constraint violation degree of the particle, and based on the feasibility rules of the constraint violation degree.

[0101] In this embodiment, for each particle in the particle population, the continuous encoding of the discrete variable in the particle needs to be first converted into an actual control instruction recognizable by the power grid. This conversion process needs to strictly follow the operation specifications of the power grid equipment and the actual value of the discrete variable, to ensure that the converted instruction is accurate and feasible. For example, for the transformer ratio, a discrete variable, the continuous encoding needs to be mapped to the nearest legal value according to the predefined discrete value set of the transformer ratio, to generate an effective transformer tap adjustment instruction.

[0102] In a specific implementation, the application does not use traditional binary encoding for discrete decision variables (such as the capacitor reactive power compensation amount of a node, the transformer ratio, etc.), but innovatively uses the "continuous encoding + probability mapping" method.

[0103] Specifically, within the optimization algorithm, the discrete decision variable is still represented by a continuous value and participates in the optimization. When the penalty function of a particle needs to be calculated, the continuous value is mapped to the actual discrete value. The mapping rule is not simply rounding off, but uses a probability value. Assuming that a decision variable in particle i is a discrete variable. In the optimization process, the is treated as a continuous variable to participate in iteration, and the does not satisfy the discrete constraint during iteration. To make the satisfy the discrete constraint, it needs to be rounded off, which can be rounded up or rounded down.

[0104] The variable value rounded up is:

[0105]

[0106] The variable value rounded down is:

[0107]

[0108] The probability of rounding up is:

[0109]

[0110] The corresponding down-rounded probability is:

[0111]

[0112] The up-rounding and down-rounding are two opposite events, and the application can select a rounding method according to the selected probability and Randomly select a rounding method, and assign the corresponding variable value to the discrete decision variable as the current value of the discrete decision variable.

[0113] In the later iteration, as the particle population converges, the fluctuation of the above probability distribution becomes smaller, and the above probability distribution tends to a certain fixed value, thereby realizing the natural transition from exploration to stability, and adaptively adjusting the rounding probability. This method maintains the information of the neighborhood of the discrete variable, and avoids the non-differentiable problem caused by simple rounding.

[0114] Subsequently, the application can use the converted actual control instruction and the continuous coding of the continuous variable in the particle to perform detailed power flow calculation. Power flow calculation is a basic link in power system analysis, which can simulate the power distribution and voltage level of the power grid under a certain operating state. The application can bring the continuous coding of the continuous variable in each particle and the actual control instruction into the above active and reactive power flow equations of the power grid. Through power flow calculation, the voltage amplitude and phase angle of each node, the power flow of each branch, and other key information can be obtained, which are crucial for evaluating the fitness and constraint violation degree of the particle.

[0115] After obtaining the power flow calculation results, the application can determine the fitness and constraint violation degree of the particle based on the results. Fitness is an important indicator for measuring the quality of the particle solution, which is usually inversely proportional to the network loss, i.e., the smaller the network loss, the higher the fitness. The constraint violation degree is used to quantify the degree of violation of the given constraint conditions by the particle solution, such as node voltage out-of-limit, line power overload, etc. By calculating the fitness and constraint violation degree, the advantages and disadvantages of the particle can be comprehensively evaluated.

[0116] Finally, according to the fitness and constraint violation degree of the particle, and based on the feasibility rules of the constraint violation degree, the historical optimal position of the particle and the global optimal position of the particle population are updated. The feasibility rule is an effective optimization strategy, which can gradually guide the particle to move to a more optimal region while ensuring the feasibility of the solution.

[0117] Specifically, for the particle with higher fitness and lower constraint violation, the historical optimal position of the particle is updated to the current position; meanwhile, among the historical optimal positions of all particles, the position with the highest fitness and the lowest constraint violation is selected as the global optimal position of the particle population. Through this process, it can be ensured that the particle population constantly approaches the global optimal solution in the optimization process.

[0118] In an embodiment, the constraint conditions can include equality constraints and inequality constraints.

[0119] In S1413, the fitness and the constraint violation of the particle are determined according to the power flow calculation result, which can include:

[0120] In S14131, the power flow residual that is not completely satisfied in the power flow calculation process is obtained.

[0121] In S14132, the constraint violation of the particle is calculated according to the power flow residual and the total amount of over-limit of all inequality constraints.

[0122] In S14133, the total active network loss of the whole network is calculated according to the power flow calculation result and the objective function.

[0123] In S14134, the fitness of the particle is calculated based on the total active network loss of the whole network, the first penalty term corresponding to the inequality constraint, and the second penalty term corresponding to the equality constraint.

[0124] In this embodiment, when calculating the constraint violation and the fitness of the particle, the power flow residual that is not completely satisfied in the power flow calculation process can be obtained first. The power flow residual is an important indicator for measuring the gap between the power flow calculation result and the ideal state, which reflects the degree of power imbalance of each node in the power grid under a specific operating state. The power flow residual that is not completely satisfied means that there is power loss or imbalance in the power grid, which may have an adverse impact on the stability and economy of the power grid.

[0125] Therefore, the power flow residual generated in the power flow calculation process can be obtained, specifically by comparing the difference between the injection power of each node obtained by power flow calculation and the ideal injection power, and then summarizing and analyzing these differences to obtain the value of the power flow residual that is not completely satisfied. This value can intuitively reflect the deviation between the operating state of the power grid represented by the current particle and the ideal state.

[0126] Then, the application can calculate the constraint violation degree of the particle according to the power flow residual and the sum of the over-limit quantities of all inequality constraints. Inequality constraints usually include node voltage over-limit, line power overload, etc., which ensure the safe operation of the power grid within a safe range. The sum of the over-limit quantities refers to the sum of the degrees of violation of the particle solution to these inequality constraints, which reflects the infeasibility of the particle solution. By comprehensively considering the power flow residual and the sum of the over-limit quantities, the constraint violation degree of the particle can be more comprehensively evaluated.

[0127] Meanwhile, the application can calculate the total active network loss of the entire network according to the power flow calculation results and the objective function. The objective function is usually set to minimize the total network loss of the entire network, so the total active network loss of the entire network is one of the key indicators for evaluating the pros and cons of the particle solution. Through power flow calculation, the power distribution and voltage level of the power grid under a specific operating state can be obtained, and then the total active network loss of the entire network can be calculated. For example, the application can collect the active power loss data of each branch and accumulate them to obtain the total active network loss value of the entire network. This value directly reflects the loss in the energy transmission process of the power grid operation scheme represented by the current particle, and the smaller the network loss, the more efficient the energy utilization of the operation scheme.

[0128] Finally, based on the total active network loss of the entire network, the first penalty term corresponding to the inequality constraints, and the second penalty term corresponding to the equality constraints, the fitness of the particle can be calculated. The first penalty term and the second penalty term are introduced to ensure that the particle solution satisfies the given constraints, and they correspond to the violation degrees of the inequality constraints and the equality constraints, respectively. By combining the total active network loss of the entire network with the penalty terms, the fitness of the particle can be calculated to comprehensively evaluate the pros and cons of the particle. The higher the fitness, the closer the particle solution is to the global optimal solution, and the stronger the ability to satisfy the constraints.

[0129] In a specific implementation, the large-scale power grid reactive power optimization problem of the application contains equality constraints and inequality constraints, and the objective function is to minimize the network loss. Therefore, the application can further simplify the reactive power optimization problem using the following general expression:

[0130]

[0131] s.t.

[0132]

[0133] wherein, is the decision variable; the first formula represents the objective function, the second formula represents the equality constraint, and the third formula represents the inequality constraint. p and q are the numbers of equality constraints and inequality constraints, respectively.

[0134] For the decision variable column vector The constraint violation degree BR of which is defined as:

[0135]

[0136] wherein, is the aforementioned equality constraint, is the aforementioned inequality constraint.

[0137] Further, the present application introduces a tangent function, and constructs the constraint term as a smooth transition penalty term, which is approximately linearly penalized when the constraint violation is small, keeps the gradient stable; when the constraint violation is large, the penalty is saturated, avoiding numerical overflow; when the constraint is strictly satisfied, the penalty is zero, providing accuracy guarantee for the reactive power optimization.

[0138] For the inequality constraint, a first penalty term is constructed:

[0139]

[0140] For the equality constraint, a second penalty term is constructed:

[0141]

[0142] wherein, is a coefficient for controlling the steepness of the penalty curve. For the large grid reactive power optimization problem, it is recommended to take 0.3.

[0143] The above two penalty terms are added to the objective function to obtain a penalty function:

[0144]

[0145] wherein, is a penalty factor. For the large grid reactive power optimization problem, it is recommended to take 0.5.

[0146] At this time, if the penalty function is minimized, the global optimum is obtained, which is equivalent to the reactive power optimization problem containing constraint conditions. That is, by introducing the penalty term and the penalty function, the original reactive power optimization problem containing a large number of equality and inequality constraints is converted into a penalty function optimization problem without constraint conditions.

[0147] On this basis, the present application can calculate the fitness of the particle based on the total active power loss of the whole grid, the first penalty term corresponding to the inequality constraint, and the second penalty term corresponding to the equality constraint. Specifically, the present application can weight-sum the total active power loss of the whole grid and the first and second penalty terms according to a preset weight (penalty factor) to obtain the fitness value of the particle. The smaller the fitness value, the better the particle solution meets the constraint condition while reducing the power loss, and thus the closer to the global optimal solution. Through this calculation process, the advantages and disadvantages of each particle solution can be comprehensively and accurately evaluated, providing a reliable basis for subsequent particle position updating and global optimal solution searching.

[0148] In one embodiment, updating the historical optimal position of the particle and the global optimal position of the particle population in S1414 according to the fitness and the constraint violation degree of the particle, and based on the feasibility rule of the constraint violation degree, can include:

[0149] S14141: comparing the current position, fitness and constraint violation degree of the particle with the historical optimal position, fitness and constraint violation degree of the particle, obtaining a first comparison result, and updating the historical optimal position of the particle according to the first comparison result.

[0150] S14142: comparing the historical optimal positions of all particles with each other, obtaining a second comparison result, and updating the global optimal position of the particle population according to the second comparison result.

[0151] In this embodiment, when comparing two decision variable column vectors and The application can use the following feasibility rule of the constraint violation degree:

[0152] 1. If both decision variable column vectors are feasible (BR is equal to 0), the decision variable column vector with smaller fitness value is better.

[0153] 2. If one is feasible (BR is equal to 0) and the other is infeasible (BR is greater than 0), select the feasible decision variable column vector (the decision variable column vector with BR equal to 0).

[0154] 3. If both are infeasible (BR is greater than 0), select the decision variable column vector with smaller constraint violation degree (BR).

[0155] The above decision variable column vector selection rule based on the constraint violation degree can automatically evolve the decision variable towards the boundary and interior of the feasible region, without adjusting the penalty factor, and has strong robustness.

[0156] Based on this, when updating the historical optimal position of each particle and the global optimal position of the particle population, the application can first compare the position, fitness and constraint violation degree of the current particle with the corresponding data when the historical optimal position of the particle according to the feasibility rule of the constraint violation degree. If the current particle is feasible and has better fitness, or the current particle is feasible and the historical optimal position particle is infeasible, or both are infeasible but the constraint violation degree of the current particle is smaller, then the position of the current particle is updated to the historical optimal position of the particle.

[0157] After the update of the historical optimal position of each particle, the application can compare the historical optimal positions of all particles with each other. According to the feasibility rule of constraint violation degree as described above, the position with the optimal fitness and feasibility, or the position with the minimum constraint violation degree when there is no feasible solution, is selected from the historical optimal positions of all particles as the global optimal position of the particle population. In this way, the particle population can continuously evolve towards the global optimal solution in the optimization process.

[0158] In one embodiment, the secondary update of the particle positions of all particles in the particle population based on the updated historical optimal positions of each particle and the global optimal position of the particle population, and based on the particle swarm optimization mechanism of quantum behavior and chaotic mapping, in S142 can include:

[0159] S1421: calculating the center point of the historical optimal positions of all particles according to the updated historical optimal positions of each particle.

[0160] S1422: generating a random position between the updated historical optimal positions of each particle and the global optimal position of the updated particle population for each particle.

[0161] S1423: generating a chaotic sequence using the Tent chaotic mapping formula, and determining the current contraction and expansion coefficient of each particle according to the chaotic sequence.

[0162] S1424: based on the center point, the random position of each particle, and the current contraction and expansion coefficient, performing secondary update on the particle positions of all particles in the particle population.

[0163] In this embodiment, when calculating the center point of the historical optimal positions of all particles, the application can achieve this by performing arithmetic averaging on the decision variable column vectors corresponding to the historical optimal positions of each particle. Specifically, the decision variable values on the historical optimal positions of all particles are added and divided by the total number of particles to obtain the decision variable values corresponding to the center point. This center point represents the concentration trend of the current particle population in the solution space, providing a reference benchmark for subsequent particle position update.

[0164] When generating a random position between the historical optimal position and the global optimal position for each particle, the application can use the linear interpolation method. Specifically, according to a predetermined random proportion parameter, linear combination is performed between the decision variable column vectors corresponding to the historical optimal position and the global optimal position to generate a new random position. This random position not only preserves the historical optimal information, but also introduces the guiding role of the global optimal position, which helps the particles to explore more widely in the solution space.

[0165] In the generation of chaotic sequences using the Tent chaotic mapping formula, the application can set the initial value and the number of iterations, and generate a series of pseudo-random numbers through the sensitive dependence of the chaotic system. These chaotic sequences have ergodicity and randomness characteristics, and can provide dynamic changing contraction and expansion coefficients for particle position updating. In specific implementation, the chaotic sequence can be mapped into a suitable contraction and expansion coefficient interval according to its value range, to ensure the stability and convergence of the particle updating process.

[0166] When performing the second update of the particle position based on the center point, the random position and the current contraction and expansion coefficient, the application can use the core formula of quantum behavior particle swarm optimization. Specifically, the center point is taken as the group attractor, the random position is taken as the individual cognitive guide, and the contraction and expansion coefficient is taken as the dynamic adjustment parameter, and the new position of the particle is calculated through weighted summation. This updating mechanism not only considers the group cooperation effect, but also retains the individual exploration ability, and at the same time, the dynamic coefficient introduced by the chaotic mapping enhances the ability of the algorithm to escape from the local optimum. Through multiple iterative updates, the particle population will gradually gather in the global optimal solution region.

[0167] In a specific implementation, the application fundamentally reforms the standard particle swarm optimization algorithm, introduces the concept of quantum mechanics and chaos theory, and proposes a chaotic quantum particle swarm optimization algorithm as a global exploration engine. The particles of the chaotic quantum particle swarm optimization algorithm no longer follow the deterministic trajectory of the standard particle swarm optimization algorithm, but exist in a "quantum potential field" described by a wave function, and their positions are determined through Monte Carlo random simulation. This greatly enhances the diversity and global search ability of the population.

[0168] First, the center of all particle historical optimal positions needs to be calculated:

[0169]

[0170] Where N is the number of particles; is the historical optimal position of the i-th particle in all iterations (i.e. the best value of the decision variable, which can be obtained by substituting the decision variable of the current iteration into the penalty function); (t is the iteration number) different values into the penalty function, and the size of the penalty function is compared to obtain); is the center point of all particle historical optimal positions.

[0171] Suppose the parameter , i.e. is a random number in the range of 0 to 1 following the standard normal distribution. The application can calculate the random position between the individual optimal and the global optimal as follows:

[0172]

[0173] Where, is the current global optimal position, i.e. the best value of all particles ; is the random position of the ith particle between the individual optimal and the global optimal.

[0174] The particle position is updated as follows when the particle swarm is transferred from the tth iteration to the (t+1)th iteration:

[0175]

[0176] wherein, and are the positions of the ith particle at the tth and (t+1)th iteration, respectively; is selected with equal probability; is the particle contraction and expansion coefficient. This formula enables the particle to appear at any position in the search space with a certain probability, avoiding premature convergence.

[0177] In the above process, the present application can use Tent chaotic mapping to adaptively adjust the key parameters , instead of the traditional linear decreasing strategy. The Tent mapping has better ergodicity and faster convergence speed, and is specifically as follows:

[0178]

[0179] The present application dynamically adjusts through a chaotic sequence, so that the algorithm has strong explorability in the early stage, strong development in the later stage, smooth transition, and optimization of the convergence process.

[0180] In one embodiment, based on the dimension-by-dimension convex space search algorithm and the particle positions of each particle after the secondary update in S143, the local optimal solution of the excellent particle in the particle population in different dimensions can include:

[0181] S1431: According to the fitness of each particle in the particle population, determine the excellent particle in the particle population, and determine the one-dimensional convex interval of each excellent particle in different dimensions.

[0182] S1432: Use the golden section method to search in the one-dimensional convex interval of each excellent particle in each dimension, determine the dimension value that optimizes the target function in each one-dimensional convex interval, and take the dimension value of each excellent particle in each one-dimensional convex interval as the local optimal solution of each excellent particle in each dimension.

[0183] In this embodiment, when determining the excellent particles in the particle population, the application can sort according to the fitness values calculated previously, and select several particles with the optimal fitness values as excellent particles. These excellent particles perform better in reducing network loss and meeting the constraint conditions, and therefore have more potential to find the global optimal solution. The number of excellent particles can be reasonably set according to the actual problem size and computing resources, and usually the optimal fitness particles accounting for a certain percentage (such as 10%-20%) of the total number of particles are selected as the candidate set.

[0184] When determining the one-dimensional convex interval of each excellent particle in different dimensions, the application can analyze the function value variation near the current position of each excellent particle. Specifically, for each dimension, the value of the decision variable in this dimension can be changed step by step, and the corresponding objective function value can be calculated to determine an interval that makes the objective function value convex, i.e. a one-dimensional convex interval. This interval contains the possible optimal solution, and therefore is the key area for subsequent search.

[0185] When using the golden section method for search, the application can select two points in the one-dimensional convex interval of each dimension of each excellent particle according to the golden section ratio to calculate and compare the objective function values. By continuously iterating this process, the search range can be gradually narrowed until the optimal dimension value of the objective function is found. This dimension value is the local optimal solution of the excellent particle in this dimension.

[0186] In a specific implementation, the application designs a new local search operator to address the relatively weak local development capability of the classical particle swarm algorithm. The operator searches the variables dimension by dimension (i.e. variable by variable) and converts the constraint conditions into a convex search space, greatly improving the search efficiency and feasibility.

[0187] Specifically, the dimension-by-dimension search of the application refers to searching in only one variable dimension at a time, with other dimensions fixed, which decomposes the high-dimensional problem into a series of one-dimensional problems, thereby significantly reducing the search difficulty. For example, for the current particle i, in the j-th dimension, the feasible interval of the decision variable under the current operating condition is determined according to the inequality constraint (mainly the voltage and safety domain constraint, which is approximately linearized by continuous flow calculation) . This interval is a convex set, so a convex space on the feasible region is constructed, and the optimal solution is searched in the convex space. In this one-dimensional convex interval , the golden section method is used for accurate search to find the optimal dimension value of the objective function (penalty function).

[0188]

[0189] wherein, Represents the golden ratio function, search The optimal value within the interval. The Golden Section Method is a numerical optimization method for searching the extrema of a single-peaked function based on the golden ratio. Its core principle is to select calculation points within the interval that conform to the golden ratio (Φ≈0.618), compare the function values, and gradually narrow the interval range to approach the extremum point. During the iteration process, the golden ratio property of the three-point spacing is always maintained. It belongs to the interval elimination algorithm category, along with Fibonacci search and binary search.

[0190] By performing the above operations on each dimension of the decision variable in sequence, a high-quality local optimum solution can be obtained, which can be used to improve the accuracy of excellent particles in the population.

[0191] In this way, this application can determine the local optimum for each excellent particle in different dimensions. These local optima provide important reference information for the subsequent search for the global optimum. Because the global optimum is often a combination of local optima in various dimensions, finding the local optima in each dimension can bring us closer to the global optimum. At the same time, the use of the dimensional convex space search algorithm also improves search efficiency and accuracy, enabling the algorithm to find the global optimum more quickly.

[0192] The reactive power optimization device for large power grids provided in the embodiments of this application is described below. The reactive power optimization device for large power grids described below can be referred to in correspondence with the reactive power optimization method for large power grids described above.

[0193] In one embodiment, such as Figure 3 As shown, Figure 3 This application provides a schematic diagram of a large power grid reactive power optimization device according to an embodiment of the present application. The present application also provides a large power grid reactive power optimization device, which may include a data acquisition module 210, a function construction module 220, an initialization module 230, and an optimization solution module 240, specifically including the following:

[0194] The data acquisition module 210 is used to acquire the power grid topology, line parameters, load data, generator operating limits and voltage safety limits of the target area, and to set the optimization algorithm parameters.

[0195] The function construction module 220 is used to construct an objective function and corresponding constraints that minimize the network loss of the entire network based on the power grid topology, the line parameters, the load data, the generator operating limits, and the voltage safety limits. The constraints include upper and lower limit constraints corresponding to the decision variables.

[0196] The initialization module 230 is configured to initialize a particle population randomly based on the upper and lower limit constraints of the decision variables and the optimization algorithm parameters, each particle in the particle population comprising continuous coding of all continuous variables and discrete variables, and one particle position representing one reactive power control scheme.

[0197] The optimization solving module 240 is configured to optimize and solve the particle position of each particle in the particle population based on the hybrid chaotic quantum particle swarm and dimension-by-dimension convex space search algorithm, to obtain a global optimal position that minimizes the total network loss, and take the global optimal position as the optimal control scheme during power grid dispatching.

[0198] In the above embodiment, when the power grid topology, line parameters, load data, generator operating limits, and voltage safety limits of the target region are obtained, and the optimization algorithm parameters are set, a target function and corresponding constraints that minimize the total network loss can be constructed based on the power grid topology, line parameters, load data, generator operating limits, and voltage safety limits. Since the constraints include the upper and lower limit constraints of the decision variables, the particle population can be initialized randomly based on the upper and lower limit constraints of the decision variables and the optimization algorithm parameters. Each particle in the particle population comprises continuous coding of all continuous variables and discrete variables, and one particle position represents one reactive power control scheme. Then, the particle position of each particle in the particle population can be optimized and solved based on the hybrid chaotic quantum particle swarm and dimension-by-dimension convex space search algorithm, to obtain a global optimal position that minimizes the total network loss. The global optimal position can be taken as the optimal control scheme during power grid dispatching. In the above process, the hybrid chaotic quantum particle swarm algorithm and the dimension-by-dimension convex space search algorithm are used to fully utilize the advantages of the two algorithms. The chaotic quantum particle swarm algorithm utilizes the characteristics of quantum behavior and chaotic mapping to enhance the global search ability and the ability to jump out of local optimum of the particle swarm, avoiding the premature convergence problem of traditional heuristic intelligent algorithms. The dimension-by-dimension convex space search algorithm performs fine search on different dimensions for each excellent particle to determine a local optimal solution, further improving the search accuracy and efficiency. The design of this hybrid algorithm is specifically optimized for the characteristics of the reactive power optimization problem model, such as variable property difference and constraint complexity, effectively overcoming the technical defects of poor optimization effect and efficiency of the optimization solving algorithm in the prior art.

[0199] In one embodiment, the application further provides a computer-readable storage medium having computer-readable instructions stored therein, which, when executed by one or more processors, cause the one or more processors to perform the steps of the large power grid reactive power optimization method according to any one of the above embodiments.

[0200] In one embodiment, the present application also provides a computer device, comprising: one or more processors, and a memory.

[0201] The memory stores computer readable instructions which, when executed by the one or more processors, perform the steps of the power grid reactive power optimization method of any of the above embodiments.

[0202] As shown schematically in Figure 4 , Figure 4 An internal structure diagram of a computer device provided by an embodiment of the present application is shown in FIG. 3. The computer device 300 can be provided as a server. As shown in Figure 4 , the computer device 300 includes a processing component 302, further including one or more processors, and a memory resource represented by the memory 301, for storing instructions, such as application programs, executable by the processing component 302. The application programs stored in the memory 301 can include one or more than one module each corresponding to a set of instructions. In addition, the processing component 302 is configured to execute the instructions to perform the power grid reactive power optimization method of any of the above embodiments.

[0203] The computer device 300 can also include a power component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 can operate based on an operating system stored in the memory 301, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, Free BSD TM, or the like.

[0204] Those skilled in the art can understand that Figure 4 the structure shown in FIG. 3 is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0205] Finally, it should be noted that, in the description above, relational terms such as first and second, and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0206] The various embodiments in the specification are described in progressive order with reference to numbers, each embodiment highlighting differences from other embodiments, and the various embodiments can be combined as desired, and the same or similar parts refer to each other.

[0207] The above description of disclosed embodiments provides enabling disclosure sufficient for one of ordinary skill in the art to practice or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for reactive power optimization of a large power grid, characterized by, The method comprises: acquiring power grid topology, line parameters, load data, generator operating limits and voltage safety limits of a target area, and setting optimization algorithm parameters; constructing an objective function for minimizing total network loss and corresponding constraint conditions according to the power grid topology, the line parameters, the load data, the generator operating limits and the voltage safety limits, wherein the constraint conditions comprise upper and lower limit constraints corresponding to decision variables; randomly initializing a particle population based on the upper and lower limit constraints corresponding to the decision variables and the optimization algorithm parameters, wherein each particle in the particle population comprises continuous encoding of all continuous variables and discrete variables, and a particle position represents a reactive power control scheme; optimizing and solving the particle position of each particle in the particle population based on a hybrid chaotic quantum particle swarm and a dimension-by-dimension convex space search algorithm to obtain a global optimal position for minimizing total network loss, and taking the global optimal position as an optimal control scheme during power grid dispatching.

2. The method for reactive power optimization of a large power grid according to claim 1, characterized in that, The optimization and solving of the particle position of each particle in the particle population based on the hybrid chaotic quantum particle swarm and the dimension-by-dimension convex space search algorithm to obtain a global optimal position for minimizing total network loss comprises: updating the historical optimal position of each particle in the particle population and the global optimal position of the particle population based on a feasibility rule of constraint violation degree; updating the particle position of all particles in the particle population again based on a particle swarm optimization mechanism of quantum behavior and chaotic mapping according to the updated historical optimal position of each particle and the global optimal position of the particle population; determining the local optimal solution of excellent particles in the particle population in different dimensions based on the dimension-by-dimension convex space search algorithm and the particle position of each particle after the second update, and then returning to perform the feasibility rule of constraint violation degree to update the historical optimal position of each particle in the particle population and the global optimal position of the particle population and subsequent steps until the optimization algorithm parameters are met, so as to obtain the global optimal position for minimizing total network loss.

3. The method for reactive power optimization of a large power grid according to claim 2, characterized in that, The updating of the historical optimal position of each particle in the particle population and the global optimal position of the particle population based on the feasibility rule of constraint violation degree comprises: for each particle in the particle population, converting the continuous encoding of discrete variables in the particle into actual control instructions of the power grid; performing power flow calculation according to the continuous encoding of continuous variables in the particle and the actual control instructions to obtain a power flow calculation result; determining the fitness and constraint violation degree of the particle according to the power flow calculation result; updating the historical optimal position of the particle and the global optimal position of the particle population according to the fitness and constraint violation degree of the particle and based on the feasibility rule of constraint violation degree.

4. The method of claim 3, wherein, The constraint conditions comprise equality constraints and inequality constraints. The determination of the fitness and constraint violation degree of the particle according to the power flow calculation result comprises: obtaining a power flow residual that is not completely satisfied in the power flow calculation process; calculating the constraint violation degree of the particle according to the power flow residual and the total amount of over-limit of all inequality constraints. According to the tidal current calculation result and the target function, total active network loss of the whole network is calculated; Based on the total active network loss of the whole network, the first penalty term corresponding to the inequality constraint and the second penalty term corresponding to the equality constraint, the fitness of the particle is calculated.

5. The method for reactive power optimization of a large power grid according to claim 3, characterized in that, The historical optimal position of the particle and the global optimal position of the particle population are updated according to the fitness and the constraint violation degree of the particle, and based on the feasibility rule of the constraint violation degree, including: According to the feasibility rule of the constraint violation degree, the current position, fitness and constraint violation degree of the particle are compared with the historical optimal position, fitness and constraint violation degree of the historical optimal position of the particle, and a first comparison result is obtained, and the historical optimal position of the particle is updated according to the first comparison result; According to the feasibility rule of the constraint violation degree, the historical optimal positions of all particles are compared with each other, a second comparison result is obtained, and the global optimal position of the particle population is updated according to the second comparison result.

6. The method for large power grid reactive power optimization according to claim 2, characterized in that, According to the historical optimal position of each particle and the global optimal position of the particle population after updating, and based on the particle swarm optimization mechanism of quantum behavior and chaotic mapping, the particle position of all particles in the particle population is updated again, including: According to the historical optimal position of each particle after updating, the center point of the historical optimal positions of all particles is calculated; For each particle, a random position between the historical optimal positions of each particle after updating and the global optimal position of the particle population after updating is generated; A chaotic sequence is generated using the Tent chaotic mapping formula, and the current contraction and expansion coefficient of each particle is determined according to the chaotic sequence; Based on the center point, the random position of each particle and the current contraction and expansion coefficient, the particle position of all particles in the particle population is updated again.

7. The method for large power grid reactive power optimization according to claim 2, characterized in that, Based on the dimension-by-dimension convex space search algorithm and the particle position of each particle after the second update, the local optimal solution of the excellent particle in the particle population in different dimensions is determined, including: According to the fitness of each particle in the particle population, the excellent particle in the particle population is determined, and the one-dimensional convex interval of each excellent particle in different dimensions is determined; The golden section method is used to search in the one-dimensional convex interval of each excellent particle in each dimension to determine the dimension value that optimizes the target function in each one-dimensional convex interval, and the dimension value of each excellent particle in each one-dimensional convex interval is taken as the local optimal solution of each excellent particle in each dimension.

8. A large power grid reactive power optimization device, characterized by, Including: The data acquisition module is configured to acquire the power grid topology, line parameters, load data, generator operating limits and voltage safety limits of the target region, and set optimization algorithm parameters; The function construction module is configured to construct a target function and corresponding constraint conditions for minimizing the total network loss according to the power grid topology, line parameters, load data, generator operating limits and voltage safety limits, wherein the constraint conditions include upper and lower limit constraints corresponding to decision variables. An initialization module is configured to initialize a particle population randomly based on upper and lower limit constraints corresponding to the decision variables and the optimization algorithm parameters, wherein each particle in the particle population comprises continuous coding of all continuous variables and discrete variables, and one particle position represents one reactive power control scheme; An optimization solving module is configured to optimize and solve the particle position of each particle in the particle population based on a hybrid chaotic quantum particle swarm and dimension-by-dimension convex space search algorithm, to obtain a global optimal position that minimizes the total network loss, and to take the global optimal position as an optimal control scheme for power grid dispatching.

9. A computer-readable storage medium, characterized in that: The computer readable storage medium stores computer readable instructions, and the computer readable instructions, when executed by one or more processors, cause the one or more processors to perform the steps of the method for reactive power optimization of a large power grid according to any one of claims 1 to 7.

10. A computer device, comprising: Comprise: One or more processors, and a memory; The memory stores computer readable instructions, and the computer readable instructions, when executed by the one or more processors, perform the steps of the method for reactive power optimization of a large power grid according to any one of claims 1 to 7.

Images