Power dispatching model solution optimization system and method

Abstract

The invention belongs to the field of power system automation, and discloses a power dispatching model solution optimization system and method, and the system comprises a bottleneck recognition module which is used for recognizing the solution bottleneck of a power dispatching model, and generating a solution optimization strategy according to the solution bottleneck; the mixed integer programming acceleration module is used for processing a mixed integer programming model in the power dispatching model to ensure rapid convergence and high-precision solution; the random mixed integer programming acceleration module is used for performing multi-processing on a random mixed integer programming model in the power dispatching model and ensuring the high efficiency and the reliability of a dispatching scheme of a solving result in a complex and uncertain environment; and the optimization engine integration module is used for modularly packaging each power dispatching model solving algorithm, establishing a unified calling interface for calling and dynamically calling computing resources in the power dispatching model solving process. The method can meet the requirements of a novel power system for high efficiency, real-time performance and reliability of power dispatching model solving, and ensures stable operation of the power system.

Description

Technical Field

[0001] The present invention belongs to the field of power system automation and relates to a power dispatching model solving and optimizing system and method. Background Art

[0002] With the global energy transition and large-scale access of renewable energy, the dispatching operation of power systems is facing unprecedented complexity and challenges. Traditional power dispatching methods are mainly based on deterministic optimization theory. However, in a new type of power system, affected by the unstable output of new energy, fluctuating load demand, and the spatio-temporal coupling characteristics of power systems, the dispatching problem presents higher complexity and uncertainty.

[0003] Power dispatching optimization mainly includes problems such as economic dispatching, unit commitment, and power flow optimization. These problems are usually solved by methods such as linear programming (LP), non-linear programming (NLP), and mixed integer programming (MIP). However, traditional optimization methods have the following limitations: low computational efficiency: In the face of large-scale mixed integer programming models, the solution speed of traditional methods is difficult to meet the requirements of real-time dispatching of actual power systems. Single structure: Most methods do not fully consider the complex spatio-temporal coupling structure of the power grid and cannot efficiently handle the coupling constraints and variables in the dispatching problem. Insufficient handling of uncertainty: Traditional optimization methods rely on deterministic models and lack effective means to handle the randomness and uncertainty brought by the access of new energy, affecting the reliability of dispatching schemes.

[0004] In a new type of power system, the output of renewable energy such as wind power and photovoltaic power has significant randomness and intermittency, leading to huge challenges in the supply-demand balance of power systems. For example, the output of renewable energy is difficult to accurately predict, increasing the uncertainty of dispatching schemes, and the rapid fluctuations in the output of new energy pose higher requirements for the frequency stability of the power grid. At the same time, with the wide application of electric vehicles, energy storage devices, and distributed energy, power loads show diverse and random characteristics. For example, the uneven spatio-temporal distribution of load demand exacerbates the complexity of power grid dispatching, and the random fluctuations in electricity demand increase the difficulty of supply-demand matching. In addition, the complexity of power dispatching optimization problems is also increasing continuously. For example, it involves a large number of continuous and discrete variables, highly coupled power flow constraints, network security constraints, and generation output constraints, and the dispatching problem involves different time scales and spatial ranges. It can be seen that due to the large-scale access of renewable energy, the diversity and uncertainty of load demand, and the complexity of power dispatching optimization problems in a new type of power system, the power dispatching model has become more complex. Therefore, how to solve and optimize the power dispatching model to accelerate the solution speed of the power dispatching model and improve the quality of the optimal solution, and then ensure the stable operation of the power system has become an urgent problem to be solved. Summary of the Invention

[0005] The object of the present invention is to overcome the above-mentioned disadvantages of the prior art and provide a power dispatching model solving and optimizing system and method.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In the first aspect of the present invention, a power dispatching model solving and optimizing system is provided, including: a bottleneck identification module, which is used to analyze the variables, constraints and branch-and-bound nodes of the power dispatching model, identify the solving bottleneck of the power dispatching model, and generate a solving optimization strategy according to the solving bottleneck; a mixed integer programming acceleration module, which is used to perform variable and constraint preprocessing on the mixed integer programming model in the power dispatching model, introduce a heuristic search strategy to determine the variables to be solved first, adjust the branch order in the branch-and-bound method solving process based on the adaptive branch variable selection strategy and the reinforcement learning algorithm, parallel process the sub-problems in the branch-and-bound method solving process based on the multi-threading technology, and introduce cutting planes in the feasible region; a stochastic mixed integer programming acceleration module, which is used to perform multiple stochastic variable modeling, uncertainty constraint transformation, stochastic scenario simulation, parallel solving of stochastic scenarios on the stochastic mixed integer programming model in the power dispatching model, and adjust the branch order in the branch-and-bound method solving process based on the reinforcement learning algorithm; an optimization engine integration module, which is used to modularly package each power dispatching model solving algorithm to obtain each algorithm module, establish a unified call interface for calling, and dynamically call the computing resources in the power dispatching model solving process.

[0008] Optionally, the analysis of the variables, constraints and branch-and-bound nodes of the power dispatching model, identifying the solving bottleneck of the power dispatching model, and generating a solving optimization strategy according to the solving bottleneck includes: by analyzing the influence of variables on the solving efficiency and the distribution and density of variables, obtaining the high-influence integer variables, high-influence semi-continuous variables and high-complexity integer variables of the power dispatching model as the variable bottleneck in the solving bottleneck of the power dispatching model; analyzing the constraints of the power dispatching model to obtain the high-coupling constraints, non-linear constraints and redundant constraints of the power dispatching model as the constraint bottleneck in the solving bottleneck of the power dispatching model; analyzing the branch-and-bound nodes of the power dispatching model to obtain the invalid branch-and-bound nodes and slow-converging branch-and-bound nodes of the power dispatching model as the efficiency bottleneck in the solving bottleneck of the power dispatching model; for the variable bottleneck, relaxing the high-influence integer variables and high-influence semi-continuous variables into continuous variables and decomposing the high-complexity integer variables into several simple variables; for the constraint bottleneck, performing merging or dimensionality reduction processing on the high-coupling constraints, removing redundant constraints, and transforming the non-linear constraints into a linearized form; for the branch-and-bound node bottleneck, pruning the invalid branch-and-bound nodes through constraint relaxation, performing heuristic search using simulated annealing, tabu search or genetic algorithm, and parallelizing the search process.

[0009] Optionally, the bottleneck identification module is further configured to obtain power grid operation data and dispatching requirements, and establish a power dispatching model based on the power grid operation data and dispatching requirements.

[0010] Optionally, the variable and constraint preprocessing of the mixed-integer programming model in the power dispatching model includes: for the unconstrained variables in the mixed-integer programming model, converting them into a constrained form through variable substitution; for the redundant constraints in the mixed-integer programming model, performing constraint merging based on the equivalent constraint merging technique; the introducing of the heuristic search strategy to determine the variables to be solved first includes: introducing the heuristic search strategy to solve first the variables in the mixed-integer programming model that contribute greatly to the objective function.

[0011] Optionally, the parallel processing of sub-problems in the branch-and-bound method solving process based on the multi-threading technology includes: allocating different sub-problems to different computing threads for parallel solving; wherein, during the parallel solving process, each sub-problem executes several linear programming sub-solvers in parallel, and the computing thread with a large computational amount is allocated more thread resources than the computing thread with a small computational amount.

[0012] Optionally, the introduction of cutting planes in the feasible region includes: introducing Gomory cutting planes, covering cutting planes, power flow constraint cutting planes, and spatio-temporal characteristic cutting planes in the feasible region; and dynamically checking whether the current solution violates the cutting planes during the solving process, and if it does, dynamically adding cutting plane constraints; and generating new cutting plane constraints in real time as the search tree expands.

[0013] Optionally, the multiple random variable modeling of the stochastic mixed-integer programming model in the power dispatching model includes: adopting a two-stage stochastic programming method, wherein, in the first stage, deterministic variables are solved, and in the second stage, the probability distributions of the uncertain variables are set, and the uncertain variables are solved through stochastic scenario compensation.

[0014] Optionally, the conversion of the uncertainty constraints includes: converting the uncertainty constraints into equivalent deterministic constraints by introducing a confidence level for the uncertainty constraints.

[0015] Optionally, the stochastic scenario simulation and stochastic scenario parallel solving include: using Monte Carlo simulation to generate a number of initial stochastic scenarios, and clustering the number of initial stochastic scenarios using K-means clustering to obtain a number of typical stochastic scenarios; performing parallel solving on the number of typical stochastic scenarios, obtaining the optimal solutions of each typical stochastic scenario and performing scenario weighted merging to obtain the final global weighted optimal solution.

[0016] In the second aspect of the present invention, there is provided a method for solving and optimizing a power dispatching model based on the above-mentioned power dispatching model solving and optimizing system, including: during the process of solving the power dispatching model, analyzing the variables, constraints and branch-and-bound nodes of the power dispatching model through a bottleneck identification module, identifying the solving bottleneck of the power dispatching model, generating a solving optimization strategy according to the solving bottleneck, and optimizing the solving process according to the solving optimization strategy; during the process of solving the power dispatching model, preprocessing variables and constraints of the mixed integer programming model in the power dispatching model, introducing a heuristic search strategy to determine the variables to be solved first, adjusting the branching order in the process of solving the branch-and-bound method based on an adaptive branching variable selection strategy and a reinforcement learning algorithm, parallel processing sub-problems in the process of solving the branch-and-bound method based on multi-threading technology, and introducing cutting planes in the feasible region through a mixed integer programming acceleration module; during the process of solving the power dispatching model, performing multiple random variable modeling, uncertainty constraint transformation, random scenario simulation, parallel solving of random scenarios, and adjusting the branching order in the process of solving the branch-and-bound method based on a reinforcement learning algorithm for the stochastic mixed integer programming model in the power dispatching model through a stochastic mixed integer programming acceleration module; during the process of solving the power dispatching model, calling each algorithm module in the optimization engine integration module through a unified call interface, and dynamically calling the computing resources of the power dispatching model solving process through the optimization engine integration module.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] The power dispatching model solving optimization system of the present invention identifies the solving bottlenecks that lead to low solving efficiency or unsatisfactory results of the power dispatching model based on a bottleneck identification module, and thus takes targeted solving optimization strategies to improve the solving efficiency and result quality. Based on a mixed-integer programming acceleration module, variable and constraint preprocessing is performed on the mixed-integer programming model in the power dispatching model, a heuristic search strategy is introduced to determine the variables to be solved first, and cutting planes are introduced in the feasible region, significantly improving the solving speed of the mixed-integer programming model; at the same time, the sub-problems in the process of solving the branch and bound method are processed in parallel based on multi-threading technology, making full use of computing resources to further accelerate the overall solving process; in addition, based on an adaptive branch variable selection strategy and a reinforcement learning algorithm, the branch order in the process of solving the branch and bound method is adjusted, which can effectively reduce the scale of the search tree in the solving process and improve the search efficiency. The mixed-integer programming acceleration module is used to provide efficient solving capabilities for the mixed-integer programming model, ensuring fast convergence and high-precision solving under large-scale mixed-integer programming models. Based on a stochastic mixed-integer programming acceleration module, multiple stochastic variable modeling, transformation of uncertain constraints, stochastic scenario simulation, parallel solving of stochastic scenarios, and adjustment of the branch order in the process of solving the branch and bound method based on a reinforcement learning algorithm are performed on the stochastic mixed-integer programming model in the power dispatching model, improving the solving efficiency of large-scale stochastic mixed-integer programming models and providing the core ability to handle uncertain factors, ensuring the efficiency and reliability of the solving results in complex uncertain environments. Based on an optimization engine integration module, each power dispatching model solving algorithm is modularly encapsulated to obtain each algorithm module and a unified call interface is established for calling, supporting flexible combined calls to adapt to various dispatching requirements, and realizing the full use of computing resources and avoiding idle computing resources by dynamically calling the computing resources in the process of solving the power dispatching model. The method of the present invention can meet the requirements of the new power system for the high efficiency, real-time performance, and reliability of the power dispatching model solving, and thus effectively ensure the stable operation of the power system. Brief Description of the Drawings

[0019] Figure 1 It is a structural block diagram of the power dispatching model solving optimization system according to an embodiment of the present invention.

[0020] Figure 2 It is a flow chart of the power dispatching model solving optimization method according to an embodiment of the present invention. Detailed Embodiments

[0021] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0023] The present invention will be further described in detail below with reference to the accompanying drawings:

[0024] See Figure 1 , in an embodiment of the present invention, a power dispatching model solving and optimizing system is provided, which effectively solves the problems of low efficiency and poor quality in solving the power dispatching model.

[0025] Specifically, the power dispatching model solving and optimizing system of the present invention includes a bottleneck identification module, a mixed integer programming acceleration module, a stochastic mixed integer programming acceleration module, and an optimization engine integration module.

[0026] Among them, the bottleneck recognition module is used to analyze the variables, constraints, and branch-and-bound nodes of the power dispatch model, identify the solution bottleneck of the power dispatch model, and generate a solution optimization strategy according to the solution bottleneck; the mixed-integer programming acceleration module is used to preprocess variables and constraints of the mixed-integer programming model in the power dispatch model, introduce a heuristic search strategy to determine the variables to be solved first, adjust the branching order in the solution process of the branch-and-bound method based on the adaptive branching variable selection strategy and the reinforcement learning algorithm, parallelly process sub-problems in the solution process of the branch-and-bound method based on multi-threading technology, and introduce cutting planes in the feasible region; the stochastic mixed-integer programming acceleration module is used to perform multiple stochastic variable modeling, uncertainty constraint transformation, stochastic scenario simulation, parallel solution of stochastic scenarios, and adjust the branching order in the solution process of the branch-and-bound method based on the reinforcement learning algorithm for the stochastic mixed-integer programming model in the power dispatch model; the optimization engine integration module is used to modularly package the solution algorithms of each power dispatch model to obtain each algorithm module, establish a unified call interface for calling, and dynamically call the computing resources in the solution process of the power dispatch model.

[0027] The power dispatching model solving and optimizing system of the present invention identifies the solving bottlenecks that lead to low solving efficiency or unsatisfactory results of the power dispatching model based on the bottleneck identification module, and thus takes targeted solving optimization strategies to improve the solving efficiency and result quality. Based on the mixed-integer programming acceleration module, variable and constraint preprocessing are performed on the mixed-integer programming model in the power dispatching model, a heuristic search strategy is introduced to determine the variables to be solved first, and cutting planes are introduced in the feasible region, significantly improving the solving speed of the mixed-integer programming model; at the same time, the sub-problems in the solving process of the branch and bound method are processed in parallel based on multi-threading technology, making full use of computing resources to further accelerate the overall solving process; in addition, based on the adaptive branch variable selection strategy and the reinforcement learning algorithm, the branch order in the solving process of the branch and bound method is adjusted, which can effectively reduce the scale of the search tree in the solving process and improve the search efficiency. The mixed-integer programming acceleration module is used to provide efficient solving capabilities for the mixed-integer programming model, ensuring fast convergence and high-precision solving under large-scale mixed-integer programming models. Based on the stochastic mixed-integer programming acceleration module, multiple stochastic variable modeling, uncertainty constraint transformation, stochastic scenario simulation, parallel stochastic scenario solving, and adjustment of the branch order in the solving process of the branch and bound method based on the reinforcement learning algorithm are performed on the stochastic mixed-integer programming model in the power dispatching model, improving the solving efficiency of large-scale stochastic mixed-integer programming models and providing the core ability to handle uncertain factors, ensuring the efficiency and reliability of the solving results in complex uncertain environments. Based on the optimization engine integration module, each power dispatching model solving algorithm is modularly encapsulated to obtain each algorithm module and a unified call interface is established for calling, supporting flexible combined calls to meet various dispatching requirements, and by dynamically calling the computing resources in the power dispatching model solving process, full use of computing resources is achieved to avoid idle computing resources. The method of the present invention can meet the requirements of the new power system for the efficiency, real-time performance, and reliability of power dispatching model solving, and thus effectively ensure the stable operation of the power system.

[0028] In a possible implementation, analyzing the variables, constraints, and branch-and-bound nodes of the power dispatch model, identifying the solution bottleneck of the power dispatch model, and generating a solution optimization strategy according to the solution bottleneck includes: obtaining high-impact integer variables, high-impact semi-continuous variables, and high-complexity integer variables of the power dispatch model by analyzing the impact of variables on the solution efficiency and the distribution and density of variables, as the variable bottleneck in the solution bottleneck of the power dispatch model; analyzing the constraints of the power dispatch model to obtain high-coupling constraints, non-linear constraints, and redundant constraints of the power dispatch model, as the constraint bottleneck in the solution bottleneck of the power dispatch model; analyzing the branch-and-bound nodes of the power dispatch model to obtain invalid branch-and-bound nodes and slow-converging branch-and-bound nodes of the power dispatch model, as the efficiency bottleneck in the solution bottleneck of the power dispatch model; for the variable bottleneck, relaxing high-impact integer variables and high-impact semi-continuous variables into continuous variables and decomposing high-complexity integer variables into several simple variables; for the constraint bottleneck, merging or dimension-reducing high-coupling constraints, removing redundant constraints, and transforming non-linear constraints into linearized forms; for the branch-and-bound node bottleneck, pruning invalid branch-and-bound nodes through constraint relaxation, performing heuristic search using simulated annealing, tabu search, or genetic algorithms, and parallelizing the search process.

[0029] Explanatory, bottleneck identification aims to effectively identify the solution performance bottleneck of the power dispatch model to provide basic support for subsequent solution optimization.

[0030] The complexity of the power dispatch model increases exponentially with the increase in the number of variables and constraints. In this implementation, the identification of the solution bottleneck is achieved through three indicators: variables, constraints, and branch-and-bound nodes.

[0031] The variables of the power dispatch model are generally divided into three types of variables: continuous variables, such as power generation; discrete variables, such as unit start-stop status. Semi-continuous variables, such as special ordered set variables, represent the piecewise linear relationship of some constraints. By analyzing the impact of variables on the solution efficiency and the distribution and density of variables, high-impact integer variables, high-impact semi-continuous variables, and high-complexity integer variables are determined.

[0032] Among them, the method for identifying high-complexity integer variables includes: First, statistically analyze the distribution of integer variables to clarify the quantity, distribution characteristics of integer variables, and their impact on the model complexity, including: 1. Variable quantity statistics: Calculate the proportion of the total number of integer variables in the total number of variables. If the proportion > 30%, it indicates that the proportion of this integer variable is relatively high, which may lead to a decrease in the solution efficiency. 2. Distribution area of integer variables: Classify integer variables by function (such as start-stop state and regulating device state, etc.), count the quantity and distribution of each type of integer variable, and identify key variable groups. 3. Distribution sparsity analysis: Calculate the sparsity of integer variables through the variable-constraint matrix (column sparsity). Second, conduct variable complexity sensitivity analysis to evaluate the specific contribution of integer variables to the problem complexity and solution time. It includes: 1. Variable relaxation experiment: Relax some integer variables into continuous variables, solve the optimization problem, and evaluate the change in the solution time before and after relaxation. If the relaxation of a certain type of integer variable brings a significant reduction in time, then this type of variable is the solution bottleneck. 2. Group sensitivity of integer variables: Group integer variables by category (such as device start-stop variables, resource allocation variables, etc.), relax each group of variables separately, and record their contribution to the overall complexity. 3. Correlation analysis of variable characteristics and performance: Use correlation analysis tools to check the relationship between variable characteristics (such as whether they are sparse) and the solution time. Third, locate high-complexity integer variables, locate the integer variables that have the greatest impact on the solution complexity, and clarify their roles in the model. It includes: 1. Branch contribution analysis: During the branch and bound process, count the number of branch expansions caused by each integer variable and calculate the branch efficiency. If the branch efficiency of a certain integer variable is much higher than that of other variables, then this variable is a high-complexity integer variable.

[0033] For high-impact integer variables and high-impact semi-continuous variables, combine the functional characteristics of the variables (such as device start-stop and charge-discharge state), and use the intermediate results and performance indicators during the solution process of the power dispatch model to identify the variables that have the greatest impact on the convergence of the power dispatch model. Exemplarily, the convergence speed of the variable can be detected by testing the performance improvement effect through the variable relaxation method.

[0034] The constraint analysis of the power dispatch model includes identifying high-coupling constraints (such as constraints across time, space, or devices) and detecting non-linear constraints and redundant constraints. Among them, the identification of high-coupling constraints is carried out in the following way: Define the coupling degree as the degree of association between variables in the constraint, which is obtained by calculating the column correlation of the constraint matrix. If the column correlation is greater than the threshold, it indicates that these two columns are highly coupled, and the corresponding constraint is a high-coupling constraint.

[0035] For the branch-and-bound node analysis of the power dispatching model, the branch-and-bound log analysis method is combined with a performance visualization tool to identify the scale of the branch-and-bound search tree and key nodes. This method can efficiently identify key nodes and performance bottlenecks by analyzing the log files of the optimization solver in detail and deeply analyzing the search tree with the visualization tool. The branch-and-bound method (Branch-and-Bound, B&B) is the core algorithm for solving mixed-integer programming problems (MIP). Its performance bottlenecks are usually reflected in the following aspects of the search tree: Search tree scale: The total number of nodes is too large. Key nodes: The solution time of some nodes is significantly longer, or it leads to a sharp increase in the number of expansions. The log analysis method obtains the following core information by parsing the branch-and-bound logs generated by the optimization solver: The solution time of the root node: used to evaluate the basic complexity of the model. The number of node expansions: identify key variables or constraints that affect the scale of the search tree. The branching efficiency of each node: analyze the effectiveness of the variable branching strategy. The solution time of sub-problems: evaluate the impact of complex sub-problems on the overall solution. Combined with the performance visualization tool, the log analysis method can quickly identify the key bottleneck nodes of the search tree.

[0036] For the identified solution bottlenecks, select suitable optimization methods or techniques to improve the solution efficiency. For variable bottlenecks: 1. Variable relaxation, relax integer variables to continuous variables to simplify the solution difficulty. 2. Variable decomposition, decompose complex variables into multiple simple variables to reduce the computational complexity. For constraint bottlenecks: 1. Constraint reduction, eliminate redundant constraints to improve the sparsity of the constraint matrix. 2. Constraint approximation, transform complex non-linear constraints into linearized forms (such as the DC approximation of power flow constraints). For branch-and-bound node bottlenecks: 1. Pruning strategy: Prune invalid nodes in advance through constraint relaxation to reduce the search space. 2. Heuristic search: Use intelligent methods such as simulated annealing, tabu search, or genetic algorithms to converge quickly. 3. Parallel computing: Decompose the search process into multiple computing nodes to improve the search efficiency.

[0037] In a possible implementation, the bottleneck identification module is further configured to obtain power grid operation data and dispatching requirements, and establish a power dispatching model according to the power grid operation data and dispatching requirements.

[0038] Explanatorily, the core objective of the power dispatching model is to achieve the balance between power grid supply and demand through optimization methods to ensure safe and economic operation. Exemplarily, the mathematical modeling of the power dispatching model can be based on the following dispatching requirements:

[0039] Minimize the generation cost:

[0040]

[0041] where C i is the cost function of unit i, P i is the generation power of unit i, ai , b i and c i are preset coefficients.

[0042] The constraint conditions generally include: power balance constraint: Among them, P demand is the system load, and P loss is the network loss. Generator output constraint: Among them, and are the lower and upper limits of the power generation of generator set i respectively. Ramping constraint: The rate of change of the generator set power is restricted. Network security constraint (such as power flow constraint): Based on the power flow calculation equation, it restricts the operation safety of the power network.

[0043] Optionally, the bottleneck identification module can also be used to construct a typical problem set, including various power dispatching scenarios, such as: large-scale generator unit combination problem, high-penetration new energy access problem, and multi-time scale (short-term, real-time) dispatching optimization problem. And it is used for performance index and test verification, including setting solution performance indexes, such as calculation time: the total time required for solution, convergence accuracy: the error between the optimal solution and the feasible solution, node search efficiency: the number of iterations of sub-problems in the branch and bound process; and constructing a performance test verification platform to evaluate the improvement of the solution performance of different solution optimization strategies.

[0044] In a possible implementation manner, the preprocessing of variables and constraints for the mixed-integer programming model in the power dispatching model includes: for the unconstrained variables in the mixed-integer programming model, they are transformed into a constrained form through variable substitution; for the redundant constraints in the mixed-integer programming model, constraint merging is performed based on the equivalent constraint merging technology.

[0045] Explanatorily, the mathematical form of the mixed-integer programming (MIP) model is generally expressed as:

[0046] min f(x) = c T x

[0047]

[0048] Among them, x is the decision variable, including the integer variable x j and the continuous variable x k . A and b represent the constraint matrix and vector. c is the objective function coefficient. X is the feasible region of the variable.

[0049] The solution of the MIP model involves the optimization of the continuous space and the combinatorial search of discrete variables, and its complexity grows exponentially with the scale of variables and constraints.

[0050] Explanatory, for the implicit free variable x, i.e., the unconstrained variable, perform variable substitution into a constrained form:

[0051] x ∈ (-∞, +∞) → x' ∈ [0, +∞), x'' = x' - x

[0052] where x', x'' ≥ 0 ≥ 0, reducing the search space.

[0053] By linearly combining equivalent constraints, merging redundant constraints, and reducing the number of constraints:

[0054] a1x + a2y = b1 and

[0055] In one possible implementation, the introduction of the heuristic search strategy to determine the variables to be solved first includes: introducing the heuristic search strategy to solve the variables that contribute greatly to the objective function in the mixed-integer programming model first.

[0056] Explanatory, according to the characteristics of the power dispatch model, design heuristic search rules to solve the variables that contribute greatly to the objective function first. Among them, in the power dispatch model, the variables that contribute greatly are the variables that have a significant impact on the objective function (such as system cost and supply-demand balance). The methods for identifying these variables are as follows: Variable sensitivity analysis: Objective function sensitivity: Calculate the partial derivative or sensitivity of the variable to the objective function. The larger the sensitivity value of the variable, the greater its contribution to the objective function. Relaxation test: Perform a relaxation test on the variable and observe the change in the objective function after relaxation. If the change is significant, then the variable is a variable that contributes greatly. Dual bound value analysis: Calculate the increase in the dual bound value of the variable and preferentially select the variable with a large increase value.

[0057] In one possible implementation, based on the adaptive branch variable selection strategy and the reinforcement learning algorithm to adjust the branching order in the branch and bound method solving process. Exemplarily, the reinforcement learning algorithm can use a pre-trained deep reinforcement learning (such as DQN) model to dynamically select branch variables to adjust the branching order. Among them, the pre-trained deep reinforcement learning model trains the agent through the interaction process of state-action-reward and dynamically selects the optimal branch variable. The specific process is as follows: State definition: Use the information of the current node of the search tree as the state, including: the variable values of the current node, the scale of the remaining sub-problems, and the difference between the current objective value and the dual bound value. Action definition: The action is to select a certain variable for branching. Reward function: The reward function is defined according to the improvement of the objective function or the improvement of the search efficiency after branching. Training process: Use the DQN algorithm to optimize the following objective: where Q(S t ,x j ) represents selecting the variable x t in the state S jThe cumulative reward value. Branch order: Predict the optimal branch variable in each state and dynamically adjust the branch order.

[0058] In a possible implementation, the parallel processing of sub-problems in the branch and bound method solution based on multi-thread technology includes: allocating different sub-problems to different computing threads for parallel solution; wherein, during the parallel solution process, each sub-problem executes several linear programming sub-solvers in parallel, and the computing thread with a larger computational amount is allocated more thread resources than the computing thread with a smaller computational amount.

[0059] Explanatorily, allocate sub-problems to multiple computing threads for parallel solution to accelerate the traversal of the search tree. Among them, the linear programming sub-solvers include the primal simplex method, the dual simplex method, and the interior point method. And optimize the thread resource allocation, give priority to processing nodes with larger computational amounts, avoid excessive differences in nodes, and improve the parallel solution efficiency.

[0060] In a possible implementation, introducing cutting planes in the feasible region includes: introducing Gomory cutting planes, covering cutting planes, power flow constraint cutting planes, and spatio-temporal characteristic cutting planes in the feasible region; and dynamically checking whether the current solution violates the cutting plane during the solution process, and if it does, dynamically adding cutting plane constraints; and generating new cutting plane constraints in real time as the search tree expands.

[0061] Explanatorily, narrow the feasible region through Gomory cutting planes and covering cutting planes to improve the convergence speed of the root node. And dynamically generate specific cutting planes according to the power grid power flow constraints and spatio-temporal characteristics, namely power flow constraint cutting planes and spatio-temporal characteristic cutting planes. Among them, Gomory cutting plane: Extract the cutting plane from the simplex table of the current solution. Covering cutting plane: For the set covering constraint of the integer programming problem, generate a covering cutting plane for the variable set that exceeds the covering condition. Power flow constraint cutting plane: Generate a cutting plane according to the power flow constraint, convert the power flow imbalance part into a cutting plane constraint, and reduce infeasible solutions. Spatio-temporal characteristic cutting plane: Dynamically generate time-related constraints to limit the total amount of cross-time variables. In addition, dynamically add cutting constraints during the solution process to reduce redundant iterations, including dynamically checking whether the current solution violates the cutting plane during the solution process, and if it does, dynamically adding cutting plane constraints, and generating new cutting plane constraints in real time as the search tree expands.

[0062] In a possible implementation, the mixed integer programming acceleration module further includes quickly generating an initial feasible solution through a heuristic method to shorten the solution time. For example, heuristic initial solution: Use a heuristic algorithm to quickly generate an initial solution, such as the greedy algorithm; relaxed initial solution: Generate an integer initial solution through the solution of the continuously relaxed problem. And use an exact algorithm to verify the result to ensure the feasibility and accuracy of the solution.

[0063] In a possible implementation manner, the multi-random variable modeling of the stochastic mixed-integer programming model in the power dispatching model includes: adopting a two-stage stochastic programming method, where in the first stage, deterministic variables are solved, and in the second stage, the probability distribution of the uncertainty variables is set, and the uncertainty variables are solved through stochastic scenario compensation.

[0064] Explanatory, set the probability distribution of the uncertainty variables, such as defining the random variables of wind power, photovoltaic output, and load fluctuations, and assume that they satisfy a certain probability distribution, such as normal distribution or Beta distribution.

[0065] In a possible implementation manner, the conversion of the uncertainty constraints includes: by introducing a confidence level for the uncertainty constraints, converting the uncertainty constraints into equivalent deterministic constraints.

[0066] Explanatory, introduce a confidence level 1-α for the uncertainty constraints: P(Ax≤b(ξ))≥1-α, and convert the uncertainty constraints into equivalent deterministic constraints: Among them, is the expected value, σ is the standard deviation of the random variable, and z α is the quantile at the confidence level.

[0067] In a possible implementation manner, the stochastic scenario simulation and stochastic scenario parallel solution include: using Monte Carlo simulation to generate a number of initial stochastic scenarios, and adopting K-means clustering to cluster the number of initial stochastic scenarios to obtain a number of typical stochastic scenarios; parallelly solving the number of typical stochastic scenarios to obtain the optimal solutions of each typical stochastic scenario and performing scenario weighted combination to obtain the final global weighted optimal solution.

[0068] Explanatory, generate a large number of initial stochastic scenarios through Monte Carlo simulation, and each scenario represents the specific values of uncertain variables, such as new energy output, load demand and other variables.

[0069] Explanatory, adopting K-means clustering to cluster a number of initial stochastic scenarios includes: using the K-means clustering algorithm to cluster a large number of initial stochastic scenarios, and grouping similar initial stochastic scenarios into one group. Input: scenario data matrix, where each row represents the feature vector of a scenario. Output: K clustering centers, that is, typical stochastic scenarios, and the classification labels of each typical stochastic scenario.

[0070] Explanatory, parallelly solving the number of typical stochastic scenarios to obtain the optimal solutions of each typical stochastic scenario and performing scenario weighted combination to obtain the final global weighted optimal solution includes:

[0071] Assign weights to each typical stochastic scenario, and the weight value is determined by the occurrence frequency of this type of typical stochastic scenario in the set of initial stochastic scenarios: The optimal solutions for each typical random scenario are weighted according to their weights to obtain the global weighted optimal solution, taking into account all scenarios comprehensively.

[0072] Optionally, it is also possible to verify whether the global weighted optimal solution satisfies the global constraint conditions. If not, post-processing adjustment is performed on the result. The post-processing adjustment aims to improve the feasibility and robustness of the scheduling scheme, and the specific process is as follows: Feasibility verification: Check whether the global weighted optimal solution satisfies all constraint conditions. If a constraint is violated (such as supply-demand balance or equipment capacity limitation), enter the adjustment process. Adjustment methods: 1. Heuristic correction: For some variables that violate the constraints, use the greedy algorithm or local search to adjust the solution. 2. Lagrangian relaxation method: Introduce slack variables, relax the constraint conditions and optimize. 3. Robustness enhancement method: In response to the fluctuations of new energy and load uncertainty, increase the redundant regulation capacity. 4. Re-optimization method: Locally optimize the adjusted solution to ensure that the adjusted scheme satisfies all constraints and is close to the optimal solution.

[0073] Explanatory, the optimization engine integration module integrates linear programming, mixed integer programming, and stochastic mixed integer programming algorithms to form a unified optimization engine, realizes the service-oriented encapsulation of the algorithms, and provides flexible modular combined calls. Design a multi-threaded parallel solution framework to support node parallelism and algorithm parallelism, and improve the utilization rate of computing resources. Provide an algorithm encapsulation interface to realize data interaction and call with the power grid scheduling system. And adopt a sparse storage structure and memory optimization technology to improve the solution efficiency of the algorithm in large-scale models.

[0074] Explanatory, the computing resources for dynamically invoking the solution process of the power dispatch model include: 1. Optimize thread allocation through the task scheduler to ensure the efficient utilization of computing resources and avoid idle computing resources. 2. Reuse the already calculated solutions in the branch and bound through the warm start strategy to accelerate the solution of sub-problems and avoid repeated calculations.

[0075] See Figure 2 , in another embodiment of the present invention, a method for optimizing the solution of a power dispatch model is provided, which is implemented based on the above-mentioned power dispatch model solution optimization system.

[0076] Specifically, the power dispatching model solving optimization method includes the following steps: during the solving process of the power dispatching model, analyze the variables, constraints, and branch-and-bound nodes of the power dispatching model through the bottleneck identification module, identify the solving bottleneck of the power dispatching model, generate a solving optimization strategy according to the solving bottleneck, and optimize the solving process according to the solving optimization strategy; during the solving process of the power dispatching model, preprocess the variables and constraints of the mixed-integer programming model in the power dispatching model, introduce a heuristic search strategy to determine the variables to be solved first, adjust the branching order in the solving process of the branch-and-bound method based on the adaptive branching variable selection strategy and the reinforcement learning algorithm, parallel process the sub-problems in the solving process of the branch-and-bound method based on the multi-threading technology, and introduce cutting planes in the feasible region through the mixed-integer programming acceleration module; during the solving process of the power dispatching model, perform multiple random variable modeling, uncertainty constraint transformation, random scenario simulation, parallel solution of random scenarios, and adjust the branching order in the solving process of the branch-and-bound method based on the reinforcement learning algorithm for the stochastic mixed-integer programming model in the power dispatching model through the stochastic mixed-integer programming acceleration module; during the solving process of the power dispatching model, call each algorithm module in the optimization engine integration module through the unified call interface, and dynamically call the computing resources of the power dispatching model solving process through the optimization engine integration module.

[0077] In another embodiment of the present invention, a power dispatching system is provided, including: a dispatching data input end: used to receive the demand data of grid dispatching, including network topology, load forecasting, renewable energy output, etc.; a power dispatching model solving optimization system: model the power dispatching model and perform the above-mentioned power dispatching model solving optimization method to efficiently solve the power dispatching model; a result output end: output the optimal power dispatching scheme, including unit start-stop status, power distribution, power flow scheme, etc.

[0078] The power dispatching model solving optimization system and method of the present invention achieve high performance and high reliability in solving the power dispatching model. The specific beneficial effects are as follows:

[0079] 1. Improve the solution efficiency and break through the computational bottleneck. Through techniques such as modeling - algorithm collaboration, variable pre - processing, and constraint aggregation, pre - process large - scale power dispatch optimization problems, reducing the variable scale and the number of constraints. The adaptive branch variable selection and dynamic cutting plane generation techniques, combined with deep reinforcement learning to optimize the variable branching order, effectively reduce the size of the search tree and improve the search efficiency of the branch - and - bound method. The parallel branch - and - bound algorithm and multi - thread resource scheduling mechanism achieve parallel task solving, accelerating the overall convergence speed. The multi - linear programming solution algorithms (the primal simplex method, the dual simplex method, and the interior - point method) are executed in parallel, and the optimal method is dynamically selected to further improve the computational efficiency. Under the same problem scale, the solution speed of the present invention is significantly improved compared with traditional commercial solvers (such as CPLEX and Gurobi), and it can complete the solution of large - scale mixed - integer programming problems within the specified time, meeting the real - time requirements of power grid dispatch.

[0080] 2. Effectively handle uncertainties and improve the robustness of the solution. Through chance - constrained programming, confidence level conversion, and stochastic scenario simulation, effectively handle random factors such as new - energy output and load fluctuations. Use scenario clustering and screening techniques to simplify the computational complexity brought by random variables and retain representative stochastic scenarios. Deep reinforcement learning assists in variable selection to improve the solution efficiency of key variables under stochastic scenarios and optimize the robustness of the dispatch plan. In the case of new - energy output fluctuations and random load changes, the power dispatch plan generated by the present invention satisfies the constraint conditions at a given confidence level, with stronger robustness and reliability.

[0081] 3. Ensure system security. Get rid of the dependence on foreign commercial solvers (such as CPLEX and Gurobi), and propose an autonomous and controllable optimization engine. Through modular encapsulation and service - oriented interfaces, support the efficient solution of linear programming, mixed - integer programming, and stochastic programming. Achieve algorithm control and multi - scenario adaptation, flexibly combine algorithm modules according to the actual needs of the power system, and adapt to various power grid dispatch optimization scenarios. The present invention provides autonomous optimization and solution capabilities, solves the "black - box" problem of traditional commercial solvers and potential information security risks, and ensures the security and controllability of power dispatch optimization.

[0082] 4. Efficiently utilize resources and improve system adaptability. Through multi - thread parallel computing and task scheduling mechanisms, achieve dynamic allocation and efficient utilization of computing resources, avoiding resource waste. Design a warm - start mechanism to reduce repeated calculations and improve the solution speed of sub - problems. The algorithm modules are encapsulated through service - orientation, providing standard interfaces to support data interaction and invocation by external dispatch systems, with good scalability and compatibility. The present invention realizes the maximization of computing resource utilization, can flexibly adapt to the power dispatch optimization requirements of different scales and different scenarios, and has strong practicality and scalability.

[0083] 5. Integrate the dispatching effect to improve the economy and stability of power grid operation. Through an efficient solution algorithm, the present invention generates an optimized power dispatching plan to achieve the minimization of generation cost and the balance between power supply and demand. Under complex and uncertain conditions, the robustness of power grid dispatching is improved through stochastic optimization, ensuring the safe operation of the power grid. By modeling the uncertainty of new energy output and load forecasting, random disturbances are effectively addressed, and the adaptability of the power grid to the large-scale access of new energy is enhanced. The optimized dispatching plan reduces the system operation cost and improves the economy, stability and reliability of power grid dispatching, meeting the goals of efficient and low-carbon operation of the new power system.

[0084] 6. Adapt to the multi-scenario requirements of the new power system. The present invention is applicable to various power grid dispatching scenarios, such as large-scale unit commitment problems: through an efficient mixed-integer programming solution; high-penetration new energy scenarios: dealing with the uncertainty of the output of wind power, photovoltaic power, etc. to achieve dynamic balance between supply and demand; multi-time scale dispatching: supporting the multi-scale optimization requirements of long-term dispatching, day-ahead dispatching and real-time dispatching. It has high adaptability and flexibility and can meet the dispatching operation requirements in different scenarios of the new power system.

[0085] The power dispatching model solving and optimizing system and method of the present invention can be applied to the following scenarios:

[0086] 1. Large-scale power system dispatching operation scenario. In a large-scale power system, the dispatching center needs to coordinate the power flow among hundreds of generating units and multiple power grid regions in real time and meet the balance between supply and demand, security constraints and economic goals. However, large-scale power dispatching involves a large number of mixed-integer programming problems, including unit start-stop, power generation power allocation and power flow optimization. Traditional solution methods have bottlenecks in calculation speed and solution accuracy. Based on the power dispatching model solving and optimizing system and method, the rapid solution of large-scale unit commitment and economic dispatching problems is realized, ensuring that the dispatching center can generate the optimal plan in real time. Through variable preprocessing and adaptive branch and bound technology, the solution time is significantly reduced to meet the real-time operation requirements.

[0087] 2. High-penetration new energy power grid dispatching scenario. With the large-scale access of new energy such as wind power and photovoltaic power, the uncertain factors faced by the power system have increased significantly, including the fluctuation of new energy output and the error of load forecasting. In this case, traditional dispatching methods are difficult to effectively respond, and the dispatching plan is prone to deviate from the actual demand, affecting the safe and stable operation of the power grid. Based on the power dispatching model solving and optimizing system and method, the stochastic mixed-integer programming acceleration module fully considers the volatility of new energy output and the uncertainty of load demand through chance-constrained programming and stochastic scenario simulation to generate a highly robust dispatching plan. Through confidence level conversion, it ensures that the power grid dispatching plan meets the balance between supply and demand and security constraints under a given default probability. The adaptability of the power grid to the fluctuation of new energy output is enhanced, and the curtailment rate of wind power and photovoltaic power is reduced.

[0088] 3. Multi-time scale scheduling scenarios for regional power grids. The scheduling requirements of regional power grids are usually divided into multiple time scales such as long-term scheduling, day-ahead scheduling, and real-time scheduling. The scheduling problems at different time scales have different optimization complexities and accuracy requirements. The current scheduling system lacks a unified optimization engine and is difficult to meet the efficient solution requirements of multi-time scales. Based on the power dispatch model solving optimization system and method, different algorithm modules can be flexibly called under multi-time scales. For example, for long-term scheduling: the mixed integer programming method is used to solve the unit commitment and long-term supply-demand balance problems; for day-ahead scheduling: generate the power generation plan and network power flow scheme quickly; for real-time scheduling: achieve fast solution through parallel computing to ensure the real-time safe operation of the power grid. Thus, improve the scheduling efficiency and reliability and meet the scheduling accuracy requirements at different time scales.

[0089] 4. Economic dispatch scenarios in the electricity market environment. In the electricity market environment, power grid scheduling needs to meet both the supply-demand balance and the optimal goal of the market economy, involving multi-agent games and cost minimization problems. In addition, it is necessary to quickly complete the scheduling optimization within the market clearing time. Based on the power dispatch model solving optimization system and method, through the parallel computing framework and multi-algorithm integration, significantly accelerate the power market scheduling optimization process. By introducing deep reinforcement learning to assist decision-making, optimize the branch and bound search path, and quickly generate the optimal market clearing scheme. Improve the market clearing efficiency, ensure fair competition and optimal allocation of resources, and meet the requirements of market real-time settlement.

[0090] 5. Grid emergency dispatch scenarios under extreme weather. Under extreme weather conditions such as typhoons and blizzards, the operating state of the power grid is easily disturbed, and the uncertainty of new energy output and load demand is further increased. Traditional methods are difficult to quickly respond to emergencies and affect the stability of the power grid. Based on the power dispatch model solving optimization system and method, simulate various disturbance scenarios under extreme weather in advance and formulate emergency dispatch plans. Use confidence level conversion and reinforcement learning to assist optimization, quickly determine key variables, and improve the robustness and response speed of the dispatch plan. Ensure the safe and stable operation of the power grid under extreme conditions and reduce the risk of power supply interruption.

[0091] 6. Distributed energy and microgrid dispatch scenarios. In the context of the wide application of distributed energy, regional power grids or microgrids need to achieve coordinated and optimized dispatch of distributed energy (such as photovoltaic, wind power, energy storage), and solve the problems of random fluctuations of distributed resources and grid connection. Based on the power dispatch model solving optimization system and method, through chance-constrained programming and scenario simulation methods, effectively coordinate resources such as photovoltaic, energy storage, and load, and achieve coordinated optimization of source-network-load-storage. In the operation of the microgrid, optimize the energy storage charging and discharging strategy and distributed resource dispatch, improve the adaptability and power supply reliability of the microgrid. Reduce the energy consumption cost of the regional power grid and improve the utilization rate of new energy.

[0092] The power dispatching model solving and optimizing system and method of the present invention can be widely applied to new power systems. Especially when dealing with uncertainties and complex constraints, it has significant economy, real-time performance, and stability, providing technical support for building an intelligent, low-carbon, and reliable power system.

[0093] The division of modules in the embodiments of the present invention is illustrative, merely a logical function division. In actual implementation, there may be other division methods. In addition, in each embodiment of the present invention, the functional modules can be integrated in a processor, can also exist physically alone, or two or more modules can be integrated in one module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules.

[0094] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program code.

[0095] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of multiple flows and / or blocks.

[0096] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of multiple flows and / or blocks.

[0097] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions for implementing the steps of the function specified in one process or a plurality of processes and / or boxes Figure 1 One process or a plurality of processes and / or boxes Figure 1 The steps of the function specified in one box or a plurality of boxes

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the specific implementation manners of the present invention can still be modified or equivalently replaced, and any modification or equivalent replacement without departing from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A power dispatch model solving and optimization system, characterized in that: include: The bottleneck identification module is used to analyze the variables, constraints and branch-and-bound nodes of the power dispatch model, identify the bottleneck of the power dispatch model, and generate a solution optimization strategy based on the bottleneck; Mixed integer programming acceleration module, used to pre-process the variables and constraints of the mixed integer programming model in the power dispatching model, introduce heuristic search strategies to determine the priority variables to be solved, adjust the branch order in the branch and bound method solution process based on adaptive branch variable selection strategies and reinforcement learning algorithms, parallelly process sub-problems in the branch and bound method solution process based on multi-threading technology, and introduce cutting planes in the feasible domain; The random mixed integer programming acceleration module is used to perform multiple random variable modeling, uncertainty constraint transformation, random scenario simulation, random scenario parallel solution, and adjust the branch order in the branch and bound method solution process based on the reinforcement learning algorithm for the random mixed integer programming model in the power dispatch model; The optimization engine integration module is used to modularly encapsulate various power dispatch model solution algorithms to obtain various algorithm modules and establish a unified calling interface for calling and dynamically calling the computing resources of the power dispatch model solution process.

2. The power dispatching model solving and optimization system according to claim 1 is characterized in that: The analyzing the variables, constraints and branch and bound nodes of the power dispatching model, identifying the bottleneck of solving the power dispatching model, and generating a solution optimization strategy according to the bottleneck includes: By analyzing the influence of variables on solution efficiency and the distribution and density of variables, high-impact integer variables, high-impact semi-continuous variables and high-complexity integer variables of the power dispatch model are obtained as variable bottlenecks in the solution bottleneck of the power dispatch model. Analyze the constraints of the power dispatch model, and obtain the high coupling constraints, nonlinear constraints and redundant constraints of the power dispatch model as the constraint bottleneck in the solution bottleneck of the power dispatch model; Analyze the branch and bound nodes of the power dispatch model, obtain the invalid branch and bound nodes and the slowly convergent branch and bound nodes of the power dispatch model, and use them as the efficiency bottleneck in the bottleneck of solving the power dispatch model; To address the variable bottleneck, high-impact integer variables and high-impact semi-continuous variables are relaxed into continuous variables, and high-complexity integer variables are decomposed into several simple variables. To address constraint bottlenecks, merge or reduce the dimension of highly coupled constraints, eliminate redundant constraints, and convert nonlinear constraints into linearized forms; For branch and bound node bottlenecks, invalid branch and bound nodes are pruned by constraint relaxation, heuristic search is performed using simulated annealing, taboo search or genetic algorithm, and the search process is performed in parallel.

3. The power dispatching model solving and optimization system according to claim 1 is characterized in that: The bottleneck identification module is also used to obtain power grid operation data and dispatching requirements, and establish a power dispatching model based on the power grid operation data and dispatching requirements.

4. The power dispatching model solving and optimization system according to claim 1 is characterized in that: The variable and constraint preprocessing of the mixed integer programming model in the power dispatching model includes: for unconstrained variables in the mixed integer programming model, converting them into constrained forms through variable replacement; for redundant constraints in the mixed integer programming model, merging constraints based on equivalent constraint merging technology; The step of introducing a heuristic search strategy to determine the variables to be solved first includes: introducing a heuristic search strategy to first solve the variables that contribute most to the objective function in the mixed integer programming model.

5. The power dispatch model solving and optimization system according to claim 1, characterized in that: The sub-problems in the branch and bound method solution process based on multi-threading technology parallel processing include: Different sub-problems are assigned to different computing threads for parallel solving; wherein, during the parallel solving process, each sub-problem executes several linear programming sub-solvers in parallel, and computing threads with large computing workload are allocated more thread resources than computing threads with small computing workload.

6. The power dispatch model solving and optimization system according to claim 1, characterized in that: The introduction of cutting planes in the feasible region includes: Introduce Gomory cutting planes, covering cutting planes, flow constraint cutting planes and spatiotemporal characteristic cutting planes in the feasible domain; dynamically check whether the current solution violates the cutting plane during the solution process, and dynamically add cutting plane constraints if it does; and generate new cutting plane constraints in real time as the search tree expands.

7. The power dispatch model solving and optimization system according to claim 1, characterized in that: The multi-random variable modeling of the random mixed integer programming model in the power dispatching model includes: A two-stage stochastic programming approach is adopted, in which the first stage solves the deterministic variables, and the second stage sets the probability distribution of the uncertain variables and solves the uncertain variables through random scenario compensation.

8. The power dispatch model solving and optimization system according to claim 1, characterized in that: The uncertainty constraint transformation includes: By introducing confidence levels into uncertainty constraints, the uncertainty constraints are transformed into equivalent deterministic constraints.

9. The power dispatch model solving and optimization system according to claim 1, characterized in that: The random scenario simulation and random scenario parallel solution include: Monte Carlo simulation is used to generate several initial random scenes, and K-means clustering is used to cluster the initial random scenes to obtain several typical random scenes; Several typical random scenarios are solved in parallel to obtain the optimal solution of each typical random scenario and then weighted merge the scenarios to obtain the final global weighted optimal solution.

10. A method for solving and optimizing a power dispatching model based on the power dispatching model solving and optimizing system according to claim 1, characterized in that: include: In the process of solving the power dispatch model, the variables, constraints and branch and bound nodes of the power dispatch model are analyzed through the bottleneck identification module, the bottleneck of solving the power dispatch model is identified, and a solution optimization strategy is generated according to the bottleneck, and the solution process is optimized according to the solution optimization strategy; In the process of solving the power dispatching model, the mixed integer programming model in the power dispatching model is preprocessed with variables and constraints through the mixed integer programming acceleration module, the heuristic search strategy is introduced to determine the priority solution variables, the branch order in the branch and bound method solution process is adjusted based on the adaptive branch variable selection strategy and reinforcement learning algorithm, the sub-problems in the branch and bound method solution process are processed in parallel based on multi-threading technology, and the cutting plane is introduced in the feasible domain; In the process of solving the power dispatch model, the random mixed integer programming acceleration module is used to perform multiple random variable modeling, uncertainty constraint transformation, random scenario simulation, random scenario parallel solution, and adjust the branch order in the branch and bound method solution process based on the reinforcement learning algorithm. In the process of solving the power dispatching model, each algorithm module in the optimization engine integration module is called through a unified calling interface, and the computing resources of the power dispatching model solving process are dynamically called through the optimization engine integration module.

Images