A method, device and storage medium for designing a power grid dispatching scheme

Abstract

The application discloses a power grid dispatching scheme design method and device, equipment and a storage medium. The method comprises the following steps: clustering and structure consistency evaluation are performed on historical operation data and simulation prediction data of a power system, a historical optimal dispatching mode feature library of a mapping relationship between storage node net load and an optimal dispatching scheme is constructed; an optimal dispatching reference template is obtained through online stage matching, discrete variables are pre-locked in combination with dual space parameters, initial state correction is completed, coupling physical constraints and non-restrictive network transmission constraints are trimmed in sequence, and finally, the compressed model is solved to obtain a power system dispatching scheme. The application fuses historical law guidance and physical mechanism verification, greatly reduces the solving time on the premise of almost no loss of dispatching economy, simultaneously adapts to various solving modes such as certainty, randomness and robustness, and guarantees the time limit certainty and reliability of large-scale power grid online dispatching.

Description

Technical Field

[0001] This application relates to the field of power system optimization and dispatching technology, and in particular to a design method, apparatus, equipment and storage medium for a power grid dispatching scheme. Background Technology

[0002] With the large-scale grid connection of high proportions of renewable energy and distributed loads, the uncertainty of power system operation has significantly increased. Power system dispatch with network constraints, as a core foundational link in power system planning, operation, and electricity market clearing, directly determines the safety and economy of grid dispatch through its solution efficiency and accuracy. However, the power system dispatch problem with network constraints involves massive discrete start-up and shutdown variables and continuous power allocation variables. The deep coupling of these two types of variables leads to an exponential expansion of the decision space. Simultaneously, the system contains numerous network transmission security constraints that are unconstrained within conventional dispatch intervals. Existing methods treat these constraints indiscriminately, causing solvers to perform unnecessary numerical verifications on a large number of invalid constraints, severely hindering computational efficiency.

[0003] Currently, accelerated solution methods for large-scale power system dispatching with network constraints mainly fall into two categories. One category is model-driven methods, such as Lagrange relaxation and mixed-integer linear programming, which design acceleration strategies by mining the mathematical structure of the model. However, when facing large-scale systems and strong uncertainties, they are prone to slow convergence speeds and difficulties in constructing feasible solutions due to over-reliance on iterative convergence mechanisms. The other category is data-driven methods, which train predictive models based on historical or simulated data to guide the solution process. However, their generalization ability is limited by the scenario coverage of the training samples, and it is difficult to guarantee the physical feasibility of the dispatching scheme from a mechanistic perspective.

[0004] Therefore, how to effectively reduce the decision space and accurately eliminate invalid constraints while maintaining global optimality, and at the same time take into account computational efficiency and solution accuracy, is an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the power grid dispatching scheme design method, apparatus, equipment, and storage medium provided in this application can effectively reduce the decision space and accurately eliminate invalid constraints while maintaining global optimality, while also taking into account computational efficiency and solution accuracy. The power grid dispatching scheme design method, apparatus, equipment, and storage medium provided in this application are implemented as follows: This application provides a method for designing a power grid dispatching scheme, including: Historical operating data and simulated forecast data of the power system are acquired, and clustering and structural consistency assessment are performed on the historical operating data and simulated forecast data to obtain a historical optimal dispatch mode feature library. The historical optimal dispatch mode feature library stores the implicit mapping relationship between node net load characteristics and unit optimal dispatch scheme. Obtain load and new energy scenario instances, perform scenario matching processing between the load and new energy scenario instances and the historical best scheduling mode feature library, and obtain the unit optimal scheduling reference template. The optimal scheduling reference template of the unit and the marginal constraint value parameters of the dual space are subjected to discrete variable pre-locking processing to obtain the pre-locking result. The initial state consistency mandatory verification and correction processing is performed on the pre-locking result to obtain the locked and corrected discrete variable state. The locked and corrected discrete variable states and system physical constraints are coupled and pruned to obtain an initial simplified constraint set; The initial simplified constraint set and the system limit power flow distribution are subjected to constraint validity determination to obtain the target simplified constraint set; A power system dispatching original model with network constraints is constructed. The locked and corrected discrete variable states, the target simplified constraint set, and the power system dispatching original model with network constraints are compressed and solved to obtain the power system dispatching scheme.

[0006] In some embodiments, the acquisition of historical operating data and simulated forecast data of the power system, and the clustering and structural consistency evaluation processing of the historical operating data and simulated forecast data to obtain a feature library of historical optimal dispatch patterns, includes: The node net load is calculated and preprocessed based on the historical operating data and simulated prediction data to obtain a node net load data sample set; The node net load data sample set is subjected to unsupervised hierarchical partitioning to obtain multiple scene clusters; The structural consistency index is calculated for the optimal power system scheduling solution within each scenario cluster to obtain the scheduling scheme stability index corresponding to each scenario cluster. The stability index of the scheduling scheme is compared with the preset structural consistency threshold to obtain a feature library of historical optimal scheduling modes.

[0007] In some embodiments, obtaining load and renewable energy scenario instances involves performing scenario matching processing between the load and renewable energy scenario instances and the historical optimal scheduling mode feature library to obtain the unit optimal scheduling reference template, including: The node net load feature is extracted from the load and new energy scenario examples to obtain the online net load feature vector; The similarity calculation is performed between the online net load feature vector and the central feature vector of each scenario cluster in the historical optimal scheduling mode feature library to obtain multiple scenario matching degree values. The multiple scene matching degree values ​​are sorted in descending order and extreme value filtering is performed to obtain the target scene cluster that is closest to the online instance; The intersection of stable scheduling schemes corresponding to the target scenario cluster is extracted and processed to obtain the optimal scheduling reference template for the unit.

[0008] In some embodiments, the step of performing discrete variable pre-locking processing on the optimal scheduling reference template of the unit and the marginal constraint value parameters of the dual space to obtain a pre-locking result, and performing initial state consistency mandatory verification and correction processing on the pre-locking result to obtain the locked and corrected discrete variable states includes: The load and new energy scenario examples are subjected to single-cycle dual relaxation solution processing to obtain the dual space marginal constraint value parameters. The dual space marginal constraint value parameters are processed by the unit state locking value calculation to obtain the unit state locking value index corresponding to each discrete variable. The unit status lock value index and the unit optimal scheduling reference template are sorted in descending order to obtain a discrete variable sequence. The discrete variable sequence and the preset reduction intensity coefficient are subjected to state locking processing to obtain the pre-locking result; The initial start-up and shutdown status, minimum continuous operating time, and minimum continuous stop time of the unit are obtained. The pre-locking result is subjected to constraint verification processing with the initial start-up and shutdown status, the minimum continuous operating time, and the minimum continuous stop time to obtain the constraint violation result. The pre-locked variables corresponding to the constraint violation results are corrected and locked to obtain the locked and corrected discrete variable states.

[0009] In some embodiments, the coupling constraint pruning process performed on the locked and corrected discrete variable states and system physical constraints to obtain an initial simplified constraint set includes: The locked and corrected discrete variable states are subjected to adjacent time period start-stop state matching processing to obtain the first target unit set; All ramping constraints corresponding to the first target unit set are removed to obtain the first stage simplified constraint set; The locked and corrected discrete variable states are subjected to minimum on / off time interval homomorphic locking identification processing with the first stage simplified constraint set to obtain the second target unit set. All minimum start-up and shutdown time constraints corresponding to the second target unit set are eliminated to obtain an initial simplified constraint set.

[0010] In some embodiments, the process of determining the constraint effectiveness of the initial simplified constraint set and the system limit power flow distribution to obtain the target simplified constraint set includes: The network transmission security constraints are processed into a compact matrix representation to obtain the network constraint matrix. The row elements corresponding to each network constraint in the network constraint matrix are rearranged in descending order to obtain the rearranged network constraint matrix. The upper limit of the physical output of the units corresponding to the elements of the rearranged network constraint matrix is ​​successively accumulated to obtain the critical unit index that satisfies the total net load boundary condition; The critical unit index and network constraint boundary are analyzed to construct and determine the boundary inequality, thereby obtaining a set of redundant network constraints. The initial simplified constraint set and the redundant network constraint set are subjected to constraint elimination processing to obtain the target simplified constraint set.

[0011] In some embodiments, the construction of the original power system dispatching model with network constraints involves compressing and solving the locked and corrected discrete variable states, the target simplified constraint set, and the original power system dispatching model with network constraints to obtain a power system dispatching scheme, including: The solution requirements for power system scheduling with network constraints are subjected to type matching processing to obtain the target solution type, which includes deterministic solution, multi-scenario stochastic solution and robust optimization solution; The solution type is compressed to adapt the locked and corrected discrete variable states, the target simplified constraint set, the target solution type, and the original power system dispatch model with network constraints to obtain a dimensionality-reduced solution model. The feasibility of the dimensionality reduction solution model is verified and iteratively solved to obtain the power system dispatching scheme.

[0012] This application provides a design apparatus for a power grid dispatching scheme, comprising: The acquisition module is used to acquire historical operating data and simulated forecast data of the power system, perform clustering and structural consistency assessment on the historical operating data and simulated forecast data, and obtain a historical optimal dispatch mode feature library. The historical optimal dispatch mode feature library stores the implicit mapping relationship between node net load characteristics and unit optimal dispatch scheme. The acquisition module is also used to acquire load and new energy scenario instances, perform scenario matching processing on the load and new energy scenario instances and the historical optimal scheduling mode feature library to obtain the optimal scheduling reference template for the unit. The processing module is used to perform discrete variable pre-locking processing on the optimal scheduling reference template of the unit and the marginal constraint value parameters of the dual space to obtain the pre-locking result, and to perform initial state consistency mandatory verification and correction processing on the pre-locking result to obtain the locked and corrected discrete variable state. The processing module is also used to perform coupling constraint pruning on the locked and corrected discrete variable states and system physical constraints to obtain an initial simplified constraint set; The processing module is also used to perform constraint validity determination processing on the initial simplified constraint set and the system limit power flow distribution to obtain the target simplified constraint set; The construction module is used to construct the original power system dispatch model with network constraints, and to perform compression and solution processing on the locked and corrected discrete variable states, the target simplified constraint set, and the original power system dispatch model with network constraints to obtain the power system dispatch scheme.

[0013] The computer device provided in this application includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements the method described in this application.

[0014] The computer-readable storage medium provided in this application embodiment stores a computer program thereon, which, when executed by a processor, implements the method described in this application embodiment.

[0015] The present application provides a design method, apparatus, equipment, and storage medium for a power grid dispatching scheme. This method involves clustering and structural consistency assessment of historical operating data and simulated prediction data of the power system to construct a feature library of historical optimal dispatching patterns that stores the mapping relationship between node net load and optimal dispatching schemes. An optimal dispatching reference template is obtained through online matching. Discrete variables are pre-locked using dual-space parameters, and initial state correction is completed. Coupled physical constraints and unconstrained network transmission constraints are then sequentially trimmed. Finally, the compressed model is solved to obtain the power system dispatching scheme. This approach effectively reduces the decision space and accurately eliminates invalid constraints while maintaining global optimality, simultaneously considering computational efficiency and solution accuracy, thus solving the technical problems mentioned in the background art. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A schematic diagram illustrating the implementation process of a power grid dispatching scheme design method provided in this application embodiment; Figure 2 A schematic diagram illustrating the implementation process of obtaining a feature library of historical optimal scheduling patterns, provided in an embodiment of this application; Figure 3 A schematic diagram of the structure of a design device for a power grid dispatching scheme provided in an embodiment of this application; Figure 4 A schematic diagram of a unit status pre-locking strategy provided in an embodiment of this application; Figure 5 This is a schematic diagram of a forced verification and correction strategy provided in an embodiment of this application; Figure 6 A time consumption comparison curve is provided for an embodiment of this application; Figure 7 This is a distribution diagram showing the start-up and shutdown status of a generator unit, provided as an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] The following description of some technologies involved in the embodiments of this application is provided to aid understanding and should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, some descriptions of well-known functions and structures are omitted in the following description.

[0020] Figure 1 This is a schematic flowchart illustrating the implementation of a power grid dispatching scheme design method provided in an embodiment of this application, including steps 101 to 106. Wherein, Figure 1 This is merely one execution order shown in the embodiments of this application and does not represent the only execution order for a power grid dispatching scheme design method. Where the final result can be achieved, Figure 1 The steps shown can be performed in parallel or in reverse order.

[0021] Step 101: Obtain historical operating data and simulated forecast data of the power system, perform clustering and structural consistency evaluation on the historical operating data and simulated forecast data, and obtain the feature library of historical optimal dispatch mode.

[0022] In this embodiment, historical operating data and simulated forecast data of the power system over a continuous period are first collected. The data dimensions include load data of each node, renewable energy output data, and system network topology information. The collected raw data is preprocessed by cleaning and normalization to calculate the net load data sample of each node, and all samples are assigned to the node net load data sample set.

[0023] An unsupervised hierarchical partitioning method is used to initially cluster the node net load data sample set, dividing all samples into multiple scenario clusters. A power system dispatch solution structure consistency index is introduced to calculate the structure consistency of the optimal power system dispatch solution corresponding to all samples within each scenario cluster, resulting in a dispatch scheme stability index for each scenario cluster. This index is used to quantify the similarity between different dispatch schemes within the same cluster.

[0024] The stability index of the scheduling scheme for each scenario cluster is compared with a preset structural consistency threshold. If the stability index of a scenario cluster is greater than or equal to the preset threshold, the cluster is determined to be a stable scenario cluster. The central feature of the scenario cluster and the stable intersection of all scheduling schemes within the cluster are saved to the historical optimal scheduling mode feature library. At the same time, the samples of the cluster are removed from the original sample set. If the stability index is less than the preset threshold, the cluster is determined to be an unstable scenario cluster. The number of scenario divisions for the cluster is increased, and the clustering step is returned to re-divide the clusters until all samples in the original sample set have been processed. Finally, the historical optimal scheduling mode feature library, which stores the implicit mapping relationship between node net load characteristics and unit optimal scheduling schemes, is obtained.

[0025] Step 102: Obtain load and new energy scenario instances, perform scenario matching processing on the load and new energy scenario instances and the historical best scheduling mode feature library to obtain the optimal scheduling reference template for the unit.

[0026] In this embodiment, an online load and new energy scenario instance to be solved is obtained, and the net load features of each node corresponding to the instance are extracted to form an online net load feature vector. Using a common similarity calculation method, the similarity between the online net load feature vector and the feature vectors of the center of each scenario cluster in the historical optimal scheduling mode feature library is calculated to obtain multiple scenario matching degree values.

[0027] All scenario matching scores are sorted in descending order, and the scenario clusters with the highest matching scores are selected as the target scenario clusters. The intersection of stable scheduling schemes corresponding to the target scenario clusters is extracted and used as a reference template for optimal unit scheduling to guide this online solution.

[0028] Step 103: Perform discrete variable pre-locking processing on the optimal scheduling reference template of the unit and the marginal constraint value parameters of the dual space to obtain the pre-locking result. Perform initial state consistency mandatory verification and correction processing on the pre-locking result to obtain the locked and corrected discrete variable states.

[0029] In this embodiment, the original problem of power system scheduling with network constraints is relaxed into a series of single-cycle subproblems that ignore time coupling constraints and start-up costs. The dual problem corresponding to the single-cycle subproblem is constructed and solved to obtain the marginal constraint value parameters of the dual space. Specifically, these parameters include the Lagrange multipliers corresponding to node power balance, the Lagrange multipliers corresponding to line forward and reverse capacity limits, and the coefficients of the piecewise linear fuel cost function of the unit.

[0030] The dual problem of the single-period subproblem is constructed as follows:

[0031] Where t is the time period index, i is the unit index, and m is the load index; It is a dual function; It is an auxiliary variable; To and Related piecewise function variables; For load; and For forward and reverse capacity limits of the line; and These are the Lagrange multipliers corresponding to the forward and reverse capacity limits of the line, respectively. and These are the upper and lower limits of the unit's power generation capacity; and Let be the piecewise linear fuel cost function coefficients for unit i.

[0032] The correlation coefficient in the above model is defined as follows:

[0033] Where t is the time period index, i is the generator set index, m is the load index, and l is the line index; It is an auxiliary variable; , and These are the Lagrange multipliers corresponding to node balancing, forward line capacity limiting, and reverse line capacity limiting, respectively. and It is the power transmission distribution factor matrix related to generator sets and loads; and For forward and reverse capacity limits of the line; For load; This refers to the line's transmission capacity.

[0034] Based on the solved dual space marginal constraint value parameters and combined with the physical characteristic parameters of the units, the unit state locking value index for each unit in the system at each scheduling period is calculated. This index, in a physical sense, reflects the comprehensive impact of market price signals such as fuel marginal cost, nodal electricity price, and line congestion price on the unit's start-up and shutdown state. The larger the index value, the higher the economic-safety comprehensive certainty of the unit's start-up and shutdown state at the corresponding time period, and the higher the priority of its state pre-locking.

[0035] The formula for calculating the unit status lock value index is as follows:

[0036] Where i is the unit index; The economic standard point for the unit; and For unit i, the piecewise linear fuel cost function coefficients are given. and These are the upper and lower limits of the unit's power generation capacity.

[0037] The derivation of the formula for calculating the unit state-locked value index is based on the optimality condition of the single-cycle dual relaxation problem: in single-cycle scheduling, the start-up and shutdown decision of unit i is essentially a comparison of its marginal revenue and marginal cost. When the unit operates at minimum output, its net revenue is... When the unit operates at maximum output, its net revenue is: Where λ is the marginal electricity price at the node, ( Let be the fuel cost function for unit i.

[0038] This application employs a piecewise linear fuel cost function. Therefore, the net benefits of the two extreme operating conditions mentioned above can be simplified as follows: Minimum output net benefit:

[0039] Maximum output net benefit:

[0040] The maximum value of the two values ​​is taken as the unit status lock-in value index because this maximum value can comprehensively cover the two extreme certainties of the unit: "must operate" and "must shut down". If the maximum value is positive and the absolute value is large, it means that no matter how the unit is scheduled within the output range, operation is significantly economical, and locking it into the operating state is highly reasonable. If the maximum value is negative and the absolute value is large, it means that no matter how the unit is scheduled, operation is uneconomical, and locking it into the shutdown state is highly reasonable.

[0041] Therefore, the larger the value of this index, the stronger the economic certainty of the unit's start-up and shutdown status during the corresponding time period, the lower the probability of pre-locking this status leading to scheduling economic losses, and the higher the pre-locking priority. According to the unit status locking value index from high to low, all discrete variables in the optimal scheduling reference template are sorted to obtain a discrete variable sequence. A reduction intensity ratio (RIR) is introduced, which is used to balance solution efficiency and scheduling economy. The larger the coefficient, the more discrete variables are locked, resulting in higher solution efficiency, but a smaller economic adjustment space for the scheduling scheme.

[0042] Combination Figure 4 As shown, according to the preset reduction intensity coefficient, the variables with the highest ranking in the discrete variable sequence are selected, and the start-stop states of these variables are strictly locked to be exactly the same as those in the optimal scheduling reference template of the unit; the remaining variables in the sequence are kept as free discrete decision variables to cope with the uncertainty fluctuations in system operation, and finally the pre-locking result of discrete variables is obtained.

[0043] The Relative Integrity (RIR) is a key parameter for system dispatchers to weigh the economics and computational efficiency of solving problems. A larger RIR means more pre-locked discrete variables, resulting in higher computational efficiency but lower economic performance. Dispatchers can improve economic performance while ensuring the feasibility of power system dispatching by adjusting the RIR appropriately. Furthermore, setting the RIR is also an effective means of ensuring the feasibility of online power system dispatching examples with network constraints. Reducing the RIR will not necessarily yield a feasible solution to an online power system dispatching example; in extreme cases, it will degenerate into the original power system dispatching problem. Therefore, on the one hand, the RIR can be adjusted according to the needs of system dispatchers; on the other hand, the RIR value of the current example can be set based on the RIR values ​​of similar historical examples. Another effective method is to run multiple examples with different RIR values ​​in parallel and select the solution with the optimality gap.

[0044] Obtain the actual online initial start-up and shutdown status of all units, as well as the inherent minimum continuous operating time and minimum continuous stop time parameters for each unit. Verify the start-up and shutdown status sequence of each unit in the pre-locking results one by one to determine whether it meets the initial state restrictions and the physical constraints of minimum continuous operating time and minimum continuous stop time. Record all pre-locking variables that violate the above constraints and their corresponding time periods to form a constraint violation result.

[0045] For pre-locked variables recorded in the constraint violation results, the guiding role of the unit's optimal scheduling reference template is forcibly removed, and the variables for the corresponding time period are corrected and locked strictly according to the actual initial start-up and shutdown state of the unit. Combined with... Figure 5 As shown, assuming and These are the number of periods during which the unit must remain in continuous operation or be shut down based on its initial state. and These are the minimum continuous operation and shutdown times of the unit, respectively. If the pre-lock result is within... and If the initial state constraints are violated, the guidance of the reference template will be forcibly removed, and the variables for that period will be strictly corrected and locked according to the unit's initial state. Simultaneously, to balance the safety constraints of the unit's minimum start-up and shutdown times with the feasibility of the solution, subsequent periods affected by this correction will be... and The start and stop states within are released again as free decision variables for flexible adjustment, ultimately resulting in a locked and corrected discrete variable state that satisfies all physical constraints.

[0046] Step 104: Perform coupling constraint pruning on the locked and corrected discrete variable states and system physical constraints to obtain an initial simplified constraint set.

[0047] In this embodiment, the coupled physical constraints in the system are selectively trimmed based on the locked and corrected discrete variable states. First, units whose start-stop states are both locked to the shutdown state in two adjacent time periods are selected to form a first target unit set. The ramp-up constraints of all units in this set in the corresponding adjacent time periods are directly removed to obtain the first stage simplified constraint set.

[0048] Next, units in which all discrete variables are locked to the same start-up and shutdown state within the minimum start-up and shutdown time interval are selected to form the second target unit set. The minimum start-up and shutdown time constraint equations of all units in this set in the corresponding time period are directly eliminated to finally obtain the initial simplified constraint set.

[0049] Step 105: Perform constraint validity determination on the initial simplified constraint set and the system limit power flow distribution to obtain the target simplified constraint set.

[0050] In this embodiment, all network transmission security constraints of the system are compactly represented in matrix form to obtain a network constraint matrix. Each row of the matrix corresponds to a transmission line in the system, and each column corresponds to a generator set in the system. Each element in the matrix represents the power transmission distribution factor of the corresponding generator set to the corresponding transmission line. Simultaneously, a boundary constraint vector corresponding to the network constraint matrix is ​​constructed. This vector contains the upper limit of the transmission capacity of each transmission line and the net load information of each node.

[0051] The network transmission security constraints of the original power system can be compactly represented in matrix form as follows:

[0052] Where G is the system network transfer factor matrix; This is a boundary constraint vector that includes the upper limit of line transmission capacity and load flow. Let t represent the power generation capacity of each unit during time period t.

[0053] For each row in the network constraint matrix, i.e., the constraint row corresponding to each transmission line, all elements in that row are sorted in descending order of value. Simultaneously, the column index of each element in the original matrix is ​​recorded to ensure accurate association with the corresponding generator set later. Let... This is a rearrangement of the original column indexes, such that the rearranged elements satisfy:

[0054] in, This represents the element in row j and column k of the network transmission constraint coefficient matrix.

[0055] In power system dispatching problems, the total physical load of the system must be less than or equal to the total generating capacity, that is:

[0056] Where i is the unit index; I is the total number of units; m is the load index; and M is the total load. For load; This represents the upper limit of the unit's generating capacity.

[0057] After successive accumulation and rearrangement, each row of elements corresponds to the upper limit of the physical output of the unit. The process continues until the accumulated value exactly covers the total net load of the system for the current period, at which point a critical unit index that satisfies this boundary condition is found. When the accumulated value reaches a certain unit, and the accumulated sum is greater than or equal to the total net load of the system for the first time, the index corresponding to that unit is recorded as the critical unit index w, i.e., the following boundary condition is satisfied:

[0058] Where r is the unit index subscript; w is the critical unit index; i is the unit index; I is the total number of units; m is the load index; and M is the total load. For load; This represents the upper limit of the unit's generating capacity.

[0059] Based on the obtained critical index w, determine whether the current system satisfies the following analytical boundary inequality:

[0060] Where r is the unit index subscript; w is the critical unit index; i is the unit index; m is the load index; and M is the total load. The element in row j and column k of the network transmission constraint coefficient matrix; For load; This represents the upper limit of the unit's generating capacity. For vectors The j-th element.

[0061] If the inequality holds, it can be rigorously proven mathematically that regardless of how the system is scheduled within the current constraint space, the original j-th network transmission constraint will never reach its capacity limit; therefore, this constraint is determined to be redundant. Traversing all rows of the network constraint matrix, the above determination process is repeated, and all network transmission constraints determined to be redundant are summarized to form a set of redundant network constraints.

[0062] All network transmission security constraint equations belonging to the redundant network constraint set are removed from the initial simplified constraint set, and the remaining valid constraints are retained. Finally, the target simplified constraint set after double pruning is obtained, which is used to solve the subsequent compressed model.

[0063] Step 106: Construct the original power system dispatch model with network constraints, and perform compressed solution processing on the locked and corrected discrete variable states, the simplified objective constraint set, and the original power system dispatch model with network constraints to obtain the power system dispatch scheme.

[0064] In this embodiment of the application, a primitive model of power system dispatch with network constraints is constructed. This model aims to minimize the total operating cost of the system and includes core constraints such as power balance, unit capacity, ramping, network transmission, and start-up and shutdown time.

[0065] The objective function of the power system dispatch model aims to minimize the total operating cost of the system, and its specific expression is as follows:

[0066] in The total operating cost is represented by z, which represents the discrete variable of the start-up and shutdown states of the thermal power unit, and p represents the power generation capacity of the thermal power unit. To initiate the cost function, This is a fuel cost function.

[0067] Meanwhile, the actual operation of this system must meet the following constraints: 1. System power balance constraint, i.e., total generating power equals total load power:

[0068] 2. Physical limits of unit generating capacity:

[0069] 3. Unit ramping constraint, that is, the output change between adjacent time periods cannot exceed the unit's ramping capacity:

[0070] 4. Network transmission security constraints:

[0071] 5. Unit start-up and shutdown state variable constraints:

[0072] In the above formula, I is the total number of generator sets, M is the total number of net load nodes, t is the time period index, and l is the transmission line index. Let i be the output of unit i during time period t. Let m be the net load of node m in time period t; and These are the upper and lower limits of the unit's generating capacity, respectively. and These are the upward and downward ramp rate limits for the unit, respectively; This represents the upper limit of the transmission capacity of line l. and , respectively, represent the power transmission distribution factors of the generator set and the load node to line l; set X represents discrete variable constraints such as the minimum start / stop time that the generator set must satisfy.

[0073] Substitute the locked and corrected discrete variable states into the original model, set the locked discrete variables to constant values, and replace the constraint set in the original model with the target simplified constraint set to obtain the compressed small-scale power system dispatch model with network constraints.

[0074] The appropriate solution mode is selected to solve the compressed model based on the actual scheduling requirements: If the scheduling requirement is deterministic, the compressed model is solved after initializing the reduction strength coefficient. If the model has no feasible solution, the reduction strength coefficient is reduced and the solution is repeated until a feasible power system scheduling scheme is obtained. If the scheduling requirement considers the uncertainty of multiple scenarios, the above variable locking and constraint pruning steps are performed for each scenario. The strict intersection of the suggested locked variables under all scenarios is taken as the final locked variable, and a unified compressed model covering all scenarios is solved. If the scheduling requirement is robust, in the main problem-sub-problem iterative architecture, the main problem is solved after each round of compression and dimensionality reduction of the current representative scenario set. Then, the worst scenario is explored through sub-problems and iterated until the robust convergence condition is met, and finally a power system scheduling scheme that meets the requirements is obtained.

[0075] Among them, for the stochastic power system scheduling problem with multiple expected patterns, after introducing the reduction strategy of this application, the mathematical model of the small-scale stochastic power system scheduling problem is as follows:

[0076] Where s is the scene index and S is the total number of scenes; Let be the probability weights for scenario s; This represents the coefficient matrix and constant vector of the various physical and network constraints of the system; For discrete variables that are pre-locked, These are free discrete decision variables that are not pre-locked.

[0077] This application's embodiments accumulate historical scheduling knowledge by building a feature library of historical best scheduling patterns offline. In the online phase, it employs a dual acceleration mechanism of "decision space compression + constraint space compression," integrating the efficiency of data-driven methods with the rigor of model-driven methods. Compared to traditional methods, this application can reduce the solution time of large-scale power system scheduling problems with network constraints by more than an order of magnitude without compromising scheduling economy, while simultaneously ensuring the feasibility of the scheduling scheme from a physical mechanism perspective. Furthermore, this application is adaptable to various mainstream scheduling solution modes, including deterministic, multi-scenario stochastic, and robust optimization, meeting the time and safety requirements of online power system scheduling in different scenarios. It effectively addresses the industry pain points of low computational efficiency and difficulty in balancing accuracy and feasibility in large-scale power grid scheduling under the background of high-proportion renewable energy grid integration. Figure 6 It can be further observed that for computationally extremely challenging online perturbation samples (with numerous peaks in the original method's curve exceeding 3000 or even 6000 seconds), the proposed method consistently and smoothly suppresses the solution time to within 500 seconds. This fully demonstrates that the decision space compression and invalid constraint pruning strategy proposed in this application can effectively suppress drastic fluctuations in solution time when facing the most computationally demanding and severe online scheduling instances, ensuring the timeliness and reliability of online scheduling. Figure 6 The light-colored lines represent the solution time curves of the original full-dimensional method, while the dark-colored lines represent the solution time curves of the fusion-based dimensionality reduction method proposed in this application.

[0078] At the same time, such as Figure 7 As shown (where black represents the shutdown state and white represents the operating state), the optimal solution of the panoramic theory obtained by the conventional benchmark method is compared with that obtained by the conventional benchmark method. Figure 7 The fast solution obtained by the fusion dimensionality reduction method of this application (left side) and the method on the left side. Figure 7 (On the right side), the start-up and shutdown trajectories of the two units highly overlap. This proves that this application maintains the economy and accuracy of system scheduling very well while accurately removing massive invalid constraints and locking high-value discrete variables.

[0079] In the above Figure 1 Based on the above, this application embodiment also provides a schematic diagram of the implementation process for obtaining a feature library of historical optimal scheduling patterns. For example... Figure 2 As shown, steps 201 to 204 are included: Step 201: Perform node net load calculation and preprocessing on historical operating data and simulated prediction data to obtain a node net load data sample set.

[0080] In this embodiment, historical operating data and simulated forecast data of the power system over a continuous period are first collected. The data dimensions cover the base load data of each node, the output data of renewable energy connected to each node, and the current network topology information of the system. The collected raw data is preprocessed to remove outliers and missing values. Then, the net load data of each node is calculated by subtracting the corresponding renewable energy output from the node's base load. After standardizing the net load data of all nodes, it is incorporated into a node net load data sample set.

[0081] Step 202: Perform unsupervised hierarchical partitioning on the node net load data sample set to obtain multiple scene clusters.

[0082] In this embodiment, the node net load data sample set is subjected to unsupervised hierarchical partitioning to obtain multiple scenario clusters. An initial number of clusters is set based on the scale of the power system and the total amount of historical data. An unsupervised hierarchical partitioning algorithm is used to divide the node net load data sample set into multiple initial scenario clusters according to the similarity of the node net load distribution characteristics. Each scenario cluster contains a group of samples with similar net load distribution characteristics.

[0083] Step 203: Calculate the structural consistency index for the optimal power system scheduling solution within each scenario cluster to obtain the stability index of the scheduling scheme corresponding to each scenario cluster.

[0084] In this embodiment, for each segmented scenario cluster, the optimal power system dispatch solution corresponding to all samples within the cluster is obtained. Each optimal solution contains the start-stop status information of all generating units within the system during all dispatch periods. The total number of optimal solutions within the cluster where the start-stop status of the same generating unit is completely consistent during the same time period is counted. This number is divided by the total number of start-stop statuses of all generating units within the cluster during all time periods to obtain the dispatch scheme stability index of the scenario cluster. The higher the index value, the stronger the consistency of the dispatch scheme corresponding to different samples within the cluster.

[0085] Step 204: Compare the stability index of the scheduling scheme with the preset structural consistency threshold to obtain the feature library of the historical best scheduling mode.

[0086] In this embodiment, a structural consistency threshold is preset, which can be adjusted according to the actual scheduling accuracy requirements. The stability index of the scheduling scheme for each scenario cluster is compared one by one with the preset threshold: if the stability index of a scenario cluster is greater than or equal to the preset threshold, the scenario cluster is determined to be a stable scenario cluster, and the central net load characteristics of the scenario cluster, as well as the portion of all optimal solutions within the cluster with completely consistent start / stop states (i.e., the intersection of stable scheduling schemes), are saved to the historical optimal scheduling pattern feature library. Simultaneously, all samples of the scenario cluster are removed from the original sample set. If the stability index of a scenario cluster is less than the preset threshold, the scenario cluster is determined to be an unstable scenario cluster, the number of partitions for the scenario cluster is increased, and the unsupervised hierarchical partitioning step is returned to re-subdivide the samples within the cluster. This comparison and partitioning process is repeated until all samples in the original sample set have been processed, ultimately resulting in a historical optimal scheduling pattern feature library storing the implicit mapping relationship between node net load characteristics and the optimal scheduling scheme of the units.

[0087] This application's embodiments unify the data format through node net load calculation and preprocessing, eliminating the interference of abnormal data on subsequent clustering; the unsupervised hierarchical partitioning method can automatically identify typical power system operation scenarios without manual annotation; the structural consistency index calculation filters out scenario clusters with stable scheduling schemes, ensuring that the scheduling patterns stored in the feature library have high reference value; through threshold comparison and iterative partitioning mechanisms, it ensures that all samples can be effectively processed, and the finally constructed historical optimal scheduling pattern feature library can accurately reflect the implicit mapping relationship between node net load characteristics and unit optimal scheduling schemes, providing a reliable knowledge foundation for rapid online solutions.

[0088] In some embodiments, load and new energy scenario instances are obtained, and scenario matching processing is performed on the load and new energy scenario instances and the historical best scheduling mode feature library to obtain the optimal scheduling reference template for the unit, including: performing node net load feature extraction processing on the load and new energy scenario instances to obtain online net load feature vectors.

[0089] Specifically, the process begins by acquiring online load and renewable energy scenario instances to be solved. These instances include load forecast data for each node within the scheduling period, renewable energy output forecast data connected to each node, and the current system's network topology information. Following calculation rules identical to those used in the offline feature library construction phase, the predicted load of each node is subtracted from its predicted renewable energy output to obtain the net load data for each node in each time period within the scheduling period. The net load data for all nodes in all time periods are then arranged according to a preset dimensional order, forming an online net load feature vector that perfectly matches the dimension of the scenario cluster center feature vector in the historical best scheduling mode feature library.

[0090] Furthermore, similarity calculations are performed on the online net load feature vector and the central feature vector of each scenario cluster in the historical best scheduling mode feature library to obtain multiple scenario matching degree values.

[0091] Specifically, the central feature vectors of all scene clusters are sequentially read from the historical best scheduling mode feature library. Using any similarity calculation method among Euclidean distance, Chebyshev distance, or standard Euclidean distance, the distance between the online net load feature vector and the central feature vector of each scene cluster is calculated. The calculated distance values ​​are then converted into corresponding scene matching scores. The smaller the distance value, the higher the scene matching score, indicating that the scene cluster is more similar to the net load distribution characteristics of the online instance.

[0092] Furthermore, the matching scores of multiple scenarios are sorted in descending order and extreme value filtering is performed to obtain the target scenario cluster that is closest to the online instance.

[0093] Specifically, all scene matching scores corresponding to all scene clusters are sorted in descending order from largest to smallest. The scene cluster with the largest matching score after sorting is selected, and the scene cluster corresponding to this score is the target scene cluster that is closest to the current online load and net load distribution characteristics of new energy scene instances.

[0094] Furthermore, the intersection of stable scheduling schemes corresponding to the target scenario cluster is extracted and processed to obtain the optimal scheduling reference template for the unit.

[0095] Specifically, the intersection of stable scheduling schemes corresponding to the target scenario cluster is extracted from the historical optimal scheduling pattern feature library. This intersection consists of the portion of all historical optimal scheduling schemes within the scenario cluster where the start-up and shutdown states of the same unit are completely consistent at the same time period, including reference start-up and shutdown state information of all units in the system throughout the entire scheduling cycle. This intersection of stable scheduling schemes is used as a reference template for optimal unit scheduling to guide this online solution, and is used for subsequent discrete variable pre-locking operations.

[0096] This application ensures consistent scene matching by extracting online net load feature vectors with dimensions consistent with the offline stage; it rapidly quantifies the similarity between online instances and historical scenes through similarity calculation; it efficiently locates the most matching target scene cluster through descending sorting and extreme value filtering; and it extracts the intersection of stable scheduling schemes corresponding to the target scene cluster as a reference template. This template represents common patterns prevalent in historical optimal solutions and has extremely high reliability. This application can complete scene matching and reference template extraction within milliseconds, providing clear guidance for subsequent discrete variable pre-locking and significantly reducing the blindness of online solutions.

[0097] In some embodiments, the optimal scheduling reference template of the unit and the marginal constraint value parameters of the dual space are subjected to discrete variable pre-locking processing to obtain the pre-locking result. The pre-locking result is subjected to initial state consistency mandatory verification and correction processing to obtain the locked and corrected discrete variable states, including: performing single-cycle dual relaxation solution processing on load and new energy scenario instances to obtain the marginal constraint value parameters of the dual space.

[0098] Specifically, the original power system dispatch problem with network constraints is relaxed into a series of independent single-cycle subproblems. During the relaxation process, the time coupling constraints of generating units (including ramp-up constraints, minimum continuous operating time constraints, and minimum continuous downtime constraints) and unit start-up costs are ignored; only the power balance constraints, generating capacity constraints, and network transmission security constraints within a single cycle are retained. The dual problem corresponding to this single-cycle subproblem is constructed and solved to obtain the marginal constraint value parameters of the dual space. These parameters include the Lagrange multipliers corresponding to node power balance, the Lagrange multipliers corresponding to line forward and reverse capacity limits, and the coefficients of the piecewise linear fuel cost function for generating units.

[0099] Furthermore, the unit state locking value is calculated by processing the marginal constraint value parameters of the dual space to obtain the unit state locking value index corresponding to each discrete variable.

[0100] Specifically, based on the dual space marginal constraint value parameters obtained from the solution, and combined with the physical characteristic parameters of the units, the unit state locking value index corresponding to each unit in the system for each scheduling period is calculated. This index reflects, in a physical sense, the comprehensive influence of market price signals such as fuel marginal cost, nodal electricity price, and line congestion price on the unit's start-up and shutdown state. The larger the index value, the higher the comprehensive economic and safety certainty of the unit's start-up and shutdown state in the corresponding period, and the higher the priority of its state pre-locking.

[0101] Furthermore, the unit status lock value index and the unit optimal scheduling reference template are sorted in descending order to obtain a discrete variable sequence.

[0102] Specifically, the optimal scheduling reference template for generating units contains the reference start / stop states of all generating units in the system during all scheduling periods, with each start / stop state corresponding to a discrete decision variable. All discrete decision variables are sorted in descending order according to their corresponding generating unit state locking value index, forming a discrete variable sequence. Variables at the beginning of the sequence are high-priority pre-locked variables.

[0103] Furthermore, a state-locking process is performed on the discrete variable sequence and the preset reduction intensity coefficient to obtain the pre-locked result.

[0104] Specifically, a reduction intensity coefficient is pre-set, with a value between 0 and 1, to balance solution efficiency and scheduling economy. The larger the coefficient, the more discrete variables are pre-locked, resulting in higher solution efficiency, but also a smaller economic adjustment space for the scheduling scheme. Based on the pre-set reduction intensity coefficient, the top-ranking variables in the discrete variable sequence are selected, and their start / stop states are strictly locked to be exactly the same as those in the optimal unit scheduling reference template. The remaining variables in the sequence remain as free discrete decision variables to cope with uncertainties and fluctuations in system operation, ultimately yielding the pre-locked results of the discrete variables.

[0105] Furthermore, the initial start-up and shutdown status, minimum continuous operating time, and minimum continuous stop time of the unit are obtained. The pre-locking result is then subjected to constraint verification processing with the initial start-up and shutdown status, minimum continuous operating time, and minimum continuous stop time to obtain the constraint violation result.

[0106] Specifically, the actual online initial start-up and shutdown status of all units at the start of the scheduling cycle is obtained, along with the inherent minimum continuous running time and minimum continuous stopping time parameters for each unit. The start-up and shutdown status sequence of each unit in the pre-locking result is verified one by one to determine whether it meets the initial state restrictions and the physical constraints of minimum continuous running time and minimum continuous stopping time. All pre-locking variables that violate the above constraints and their corresponding time periods are recorded to form the constraint violation results.

[0107] Furthermore, the pre-locked variables corresponding to the constraint violation results are corrected and locked to obtain the locked and corrected discrete variable states.

[0108] Specifically, for the pre-locked variables recorded in the constraint violation results, the guiding role of the optimal scheduling reference template for the units is forcibly removed, and the variables for the corresponding time period are corrected and locked strictly according to the actual initial start-up and shutdown states of the units. At the same time, in order to balance the safety constraints of the minimum start-up and shutdown time of the units and the feasibility of the scheduling scheme, the subsequent time period affected by this correction operation is calculated, and the unit start-up and shutdown states that were originally pre-locked within this range are released back as free discrete decision variables, allowing the solver to make flexible adjustments in subsequent time periods, and finally obtaining the locked and corrected discrete variable states that satisfy all physical constraints.

[0109] The marginal constraint value parameter obtained by the single-cycle dual relaxation solution in this application can quantify the determinism of the unit's start-up and shutdown states from both economic and safety dimensions. The unit state locking value index calculated based on this parameter can scientifically determine the pre-locking priority of discrete variables, avoiding economic losses caused by random locking. By setting a reduction strength coefficient, the solution efficiency and scheduling economy can be flexibly balanced. Through the initial state consistency mandatory verification and correction, the pre-locking results that violate physical constraints can be corrected in a timely manner, ensuring that the state of discrete variables after pre-locking conforms to the actual operating characteristics and safety limitations of the unit, thus avoiding the problem of infeasible solutions in subsequent models from the source.

[0110] In some embodiments, coupling constraint pruning is performed on the locked and corrected discrete variable states and system physical constraints to obtain an initial simplified constraint set, including: performing adjacent time period start-stop state matching processing on the locked and corrected discrete variable states to obtain a first target unit set.

[0111] Specifically, the system iterates through all generating units and all adjacent time periods within the scheduling cycle, checking the corrected discrete variable status of each generating unit in each of the two adjacent time periods. It then selects all generating units whose start / stop status is locked as stopped in both adjacent time periods within a certain set of adjacent time periods. These generating units and their corresponding adjacent time period information are then aggregated to form the first target generating unit set.

[0112] Furthermore, all ramping constraints corresponding to the first target unit set are eliminated to obtain the first-stage simplified constraint set.

[0113] Specifically, since the ramp-up constraints of the units are used to limit the output variation between adjacent time periods, and the units in the first target unit set are all in a shutdown state in the corresponding adjacent time periods, their output in both time periods must be 0, and the output variation is 0, which naturally satisfies the requirements of the ramp-up constraints. Therefore, from the original physical constraint set of the system, the ramp-up constraint equations of all units in the first target unit set in the corresponding adjacent time periods are directly removed, completing the first stage of constraint pruning and obtaining the first stage simplified constraint set.

[0114] Furthermore, the locked and corrected discrete variable states are subjected to minimum start-up time interval homomorphic locking identification processing with the first stage simplified constraint set to obtain the second target unit set.

[0115] Specifically, the system iterates through all units within the system, examining the discrete variable states within each possible minimum start-up / shutdown time interval, taking into account the minimum continuous operating time and minimum continuous shutdown time parameters for each unit. If a unit's start-up / shutdown state is locked to the operating state for all time periods within a certain minimum continuous operating time interval, or its start-up / shutdown state is locked to the shutdown state for all time periods within a certain minimum continuous shutdown time interval, then the unit is determined to be in a homomorphic locked state within the corresponding interval. All units satisfying the above homomorphic locking conditions and their corresponding minimum start-up / shutdown time interval information are then aggregated to form the second target unit set.

[0116] Furthermore, all minimum start-up and shutdown time constraints corresponding to the second target unit set are eliminated to obtain an initial simplified constraint set.

[0117] Specifically, since the start-up and shutdown states of the units in the second target unit set are completely locked within the corresponding minimum start-up and shutdown time intervals, they must satisfy the constraint requirements of minimum continuous operation or minimum continuous shutdown time. Therefore, from the first stage of simplifying the constraint set, the minimum start-up and shutdown time constraint equations of all units in the second target unit set within the corresponding time intervals are directly removed, completing the coupling constraint pruning in the second stage, and finally obtaining the initial simplified constraint set.

[0118] This application's embodiments selectively trim physical constraints tightly coupled with the locked state based on the locked discrete variable states. For units that are shut down in adjacent time periods, their ramp-up constraints are naturally satisfied, and direct removal will not affect the model's correctness. For units whose states are completely locked within the minimum start-up and shutdown time intervals, their minimum start-up and shutdown time constraints are also necessarily satisfied, and they can also be safely removed. This application can significantly reduce the number of system physical constraints without losing any feasible solution space, significantly reducing the model's solution complexity. At the same time, the trimming process only requires simple logical judgments, with extremely low computational overhead, and will not increase the overall solution time.

[0119] In some embodiments, the initial simplified constraint set and the system limit power flow distribution are subjected to constraint validity determination processing to obtain the target simplified constraint set, including: performing matrix form compact representation processing on the network transmission security constraints to obtain the network constraint matrix.

[0120] Specifically, the power transmission security constraints of all network transmission lines in the system are organized according to a unified mathematical form to construct a network constraint matrix. Each row of the network constraint matrix corresponds to a transmission line in the system, and each column corresponds to a generator set in the system. Each element in the matrix represents the power transmission distribution factor of the corresponding generator set to the corresponding transmission line. Simultaneously, a boundary constraint vector corresponding to the network constraint matrix is ​​constructed. This vector contains the upper limit of the transmission capacity of each transmission line and the net load information of each node.

[0121] Furthermore, the row elements corresponding to each network constraint in the network constraint matrix are rearranged in descending order to obtain the rearranged network constraint matrix.

[0122] Specifically, for each row in the network constraint matrix—that is, the constraint row corresponding to each transmission line—all elements in that row are sorted in descending order of value. Simultaneously, the column index of each element in the original matrix is ​​recorded to ensure accurate association with the corresponding generator set later. After performing this rearrangement operation on all rows, the rearranged network constraint matrix is ​​obtained.

[0123] Furthermore, the upper limits of the physical output of the units corresponding to the elements of the rearranged network constraint matrix are successively accumulated to obtain the critical unit index that satisfies the total net load boundary conditions.

[0124] Specifically, first, the total net load of the system during the current scheduling period is calculated, which is the total load of all nodes in the system minus the total renewable energy output. Then, for each rearranged row, starting from the first element, the physical output limit of the generator unit corresponding to that element is accumulated sequentially. After accumulating the output limit of each unit, it is determined whether the current accumulated sum is greater than or equal to the total net load of the system. When the accumulated sum is greater than or equal to the total net load of the system for the first time when accumulating to a certain unit, the index corresponding to that unit is recorded as the critical unit index.

[0125] Furthermore, analytical boundary inequalities are constructed and determined for the critical unit index and network constraint boundaries to obtain a set of redundant network constraints.

[0126] Specifically, based on the obtained critical unit index, combined with the rearranged network constraint matrix elements and boundary constraint vectors, a corresponding analytical boundary inequality is constructed. If this inequality holds, it is mathematically proven that regardless of how the system schedules unit output within the current constraint space, the actual transmission power of the corresponding transmission line will not reach its capacity limit. Therefore, this network transmission constraint is determined to be a redundant constraint. The above determination process is repeated across all rows of the network constraint matrix, and all network transmission constraints determined to be redundant are summarized to form a set of redundant network constraints.

[0127] Furthermore, constraint elimination processing is performed on the initial simplified constraint set and the redundant network constraint set to obtain the target simplified constraint set.

[0128] Specifically, all network transmission security constraint equations belonging to redundant network constraint sets in the initial simplified constraint set are removed, and the remaining valid constraints are retained. Finally, the target simplified constraint set after double pruning is obtained, which is used to solve the subsequent compressed model.

[0129] This application's embodiments use a compact matrix representation of network transmission security constraints, facilitating subsequent batch processing. By rearranging row elements in descending order and progressively accumulating the unit output limits, critical units can be quickly located. An analytical boundary inequality constructed based on the critical unit index provides a rigorous mathematical proof of the redundancy of network constraints, avoiding the conservatism and misjudgment risks of traditional constraint pruning methods. This application efficiently eliminates a large number of non-constraint network constraints in the system that will never be triggered within the normal scheduling interval, further significantly compressing the model size. Simultaneously, it ensures that the pruned constraint set has a completely equivalent feasible solution space to the original constraint set, guaranteeing that the final scheduling scheme meets all network security requirements.

[0130] In some embodiments, a power system scheduling original model with network constraints is constructed, and the locked and corrected discrete variable states, the target simplified constraint set, and the power system scheduling original model with network constraints are compressed and solved to obtain a power system scheduling scheme. This includes: performing type matching processing on the solution requirements of power system scheduling with network constraints to obtain the target solution type, which includes deterministic solution, multi-scenario stochastic solution, and robust optimization solution.

[0131] Specifically, the first step is to obtain the actual demand information for this power dispatch, including the dispatch cycle, sources of system uncertainty and their handling requirements, dispatch accuracy requirements, and calculation time limits. Based on the above information, the solution type is matched: if it is a routine day-ahead or intraday dispatch, and only the optimal dispatch scheme needs to be solved for a single typical scenario, it is determined to be a deterministic solution type; if it is necessary to consider multiple possible scenarios of renewable energy output and user load fluctuations, with the goal of minimizing the weighted average operating cost of all scenarios, it is determined to be a multi-scenario stochastic solution type; if it is necessary to ensure the safe operation of the dispatch scheme under extreme conditions, with the goal of resisting all possible fluctuations within the preset uncertainty set, it is determined to be a robust optimization solution type.

[0132] Furthermore, the solution type is compressed to adapt the locked and corrected discrete variable states, the simplified objective constraint set, the objective solution type to the original power system dispatch model with network constraints, resulting in a dimensionality-reduced solution model.

[0133] Specifically, the first step is to construct a basic model of power system dispatch with network constraints. This model takes minimizing the total operating cost of the system as its core objective and includes five basic physical constraints: system power balance, unit generating capacity, unit ramping, network transmission security, and unit start-up and shutdown time. Subsequently, adaptive compression operations are performed based on different objective solution types: For deterministic solution types, the locked and corrected discrete variable states are directly substituted into the original model, the locked discrete variables are set to constant start and stop states, and the constraint set of the original model is replaced with the objective simplified constraint set to obtain a deterministic dimensionality reduction solution model; For multi-scenario stochastic solution types, for each independent scenario in the scenario set, discrete variable pre-locking and constraint pruning operations are performed to obtain the suggested locked variable set and suggested pruned constraint set for each scenario, the strict intersection of the suggested locked variables of all scenarios is calculated as the final locked variable, the common joint invalid constraint set of all scenarios is determined, the final locked variable is substituted into the multi-scenario stochastic original model and the joint invalid constraints are removed to obtain a multi-scenario dimensionality reduction solution model; For robust optimization solution types, in the main problem-subproblem iterative architecture, for the representative scenario set of the current iteration step, discrete variable pre-locking and constraint pruning operations are performed to obtain the main problem locked variable and invalid constraint set of the current iteration step, which are substituted into the robust optimization main problem model and invalid constraints are removed to obtain the dimensionality reduction main problem model of the current iteration step.

[0134] Furthermore, the feasibility of the dimensionality reduction solution model is verified and iteratively solved to obtain the power system dispatching scheme.

[0135] Specifically, the corresponding solution and iteration process is executed according to the target solution type: If it is a deterministic solution type, the deterministic dimensionality reduction solution model is solved after initializing the reduction intensity coefficient. If a feasible solution is obtained, it is output as the final power system dispatch scheme. If the model has no feasible solution, the reduction intensity coefficient is reduced, and the discrete variable pre-locking and constraint pruning steps are returned to regenerate the dimensionality reduction model until a feasible solution that meets the requirements is obtained. If it is a multi-scenario stochastic solution type, the multi-scenario dimensionality reduction solution model is solved. If a feasible solution is obtained, it is output as the power system dispatch scheme. If the model has no solution due to the influence of extreme scenarios, the reduction intensity coefficient is reduced, the intersection of locked variables and joint invalid constraints of all scenarios are recalculated, a new dimensionality reduction model is generated, and it is solved again until a globally feasible solution is obtained. If it is a robust optimization solution type, first initialize a representative scenario set including initial extreme operating conditions, a reduction strength coefficient, and a system convergence tolerance gap. Solve the dimensionality-reduced main problem model of the current iteration step to obtain the optimal power system scheduling trial solution. Pass this trial solution to the subproblem. Within the preset uncertainty set, find the worst scenario that causes the most severe violation of system constraints. Determine whether the maximum constraint violation of the worst scenario is less than the preset convergence tolerance gap. If so, output the current trial solution as the final power system scheduling scheme; otherwise, add the worst scenario to the representative scenario set, return to execute the main problem compression and dimensionality reduction step, and enter the next round of main and subproblem iteration loop until the convergence condition is met.

[0136] This application's embodiments, through demand type matching, can automatically identify and adapt to three mainstream scheduling solution modes: deterministic, multi-scenario stochastic, and robust optimization, exhibiting broad engineering applicability. Adaptive compression processing is performed for different solution types, fully leveraging the advantages of dual compression of decision space and constraint space. Through feasibility verification and iterative solution mechanisms, when the compressed model encounters infeasibility, the reduction strength coefficient is automatically lowered and a dimensionality-reduced model is regenerated, ensuring that a feasible scheduling scheme is ultimately obtained. This application, while guaranteeing solution efficiency, also considers the reliability and robustness of the results, meeting the diverse needs of different scheduling scenarios in power systems.

[0137] While this application provides method operation steps as shown in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in this embodiment is merely one possible execution order among many and does not represent the only execution order. In actual device or client product execution, the method can be executed sequentially according to this embodiment or the accompanying drawings, or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0138] like Figure 3As shown in the illustration, this application also provides a power grid dispatching scheme design device 300. The device includes: The acquisition module 301 is used to acquire historical operating data and simulated forecast data of the power system, perform clustering and structural consistency assessment on the historical operating data and simulated forecast data, and obtain a historical optimal dispatch mode feature library. The historical optimal dispatch mode feature library stores the implicit mapping relationship between node net load characteristics and unit optimal dispatch scheme.

[0139] The acquisition module 301 is also used to acquire load and new energy scenario instances, perform scenario matching processing on the load and new energy scenario instances and the historical best scheduling mode feature library to obtain the optimal scheduling reference template for the unit.

[0140] The processing module 302 is used to perform discrete variable pre-locking processing on the optimal scheduling reference template of the unit and the marginal constraint value parameters of the dual space to obtain the pre-locking result, and to perform initial state consistency mandatory verification and correction processing on the pre-locking result to obtain the locked and corrected discrete variable states.

[0141] The processing module 302 is also used to perform coupling constraint pruning on the locked and corrected discrete variable states and system physical constraints to obtain an initial simplified constraint set.

[0142] The processing module 302 is also used to perform constraint validity determination on the initial simplified constraint set and the system limit power flow distribution to obtain the target simplified constraint set.

[0143] Module 303 is used to construct the original power system dispatch model with network constraints. It performs compressed solution processing on the locked and corrected discrete variable states, the simplified objective constraint set, and the original power system dispatch model with network constraints to obtain the power system dispatch scheme.

[0144] Some modules in the apparatus described in this application can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0145] The apparatus or module described in the above embodiments can be implemented by a computer chip or physical entity, or by a product with a certain function. For ease of description, the above apparatus is described by dividing it into various modules according to their functions. When implementing the embodiments of this application, the functions of each module can be implemented in one or more software and / or hardware. Of course, a module that implements a certain function can also be implemented by combining multiple sub-modules or sub-units.

[0146] The methods, apparatus, or modules described in this application can be implemented in a computer-readable program code manner. The controller can be implemented in any suitable manner, such as a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of a memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code manner, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included within it for implementing various functions can also be considered as structures within the hardware component. Alternatively, the device used to implement various functions can be viewed as either a software module that implements the method or a structure within a hardware component.

[0147] This application also provides an apparatus, the apparatus comprising: a processor; a memory for storing processor-executable instructions; wherein, when the processor executes the executable instructions, it implements the method described in this application.

[0148] This application also provides a non-volatile computer-readable storage medium storing a computer program or instructions thereon, which, when executed, enables the method described in this application embodiment to be implemented.

[0149] Furthermore, in the various embodiments of the present invention, each functional module can be integrated into a processing module, or each module can exist independently, or two or more modules can be integrated into a single module.

[0150] The aforementioned storage media include, but are not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Cache, Hard Disk Drive (HDD), or Memory Card. The memory can be used to store computer program instructions.

[0151] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary hardware. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, or it can be embodied in the process of data migration. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, mobile terminal, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0152] The various embodiments described in this specification are presented in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. All or part of this application can be used in numerous general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, mobile communication terminals, multiprocessor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0153] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of this application.

Claims

1. A method for designing a power grid dispatching scheme, characterized in that, include: Historical operating data and simulated forecast data of the power system are acquired, and clustering and structural consistency assessment are performed on the historical operating data and simulated forecast data to obtain a historical optimal dispatch mode feature library. The historical optimal dispatch mode feature library stores the implicit mapping relationship between node net load characteristics and unit optimal dispatch scheme. Obtain load and new energy scenario instances, perform scenario matching processing between the load and new energy scenario instances and the historical best scheduling mode feature library, and obtain the unit optimal scheduling reference template. The optimal scheduling reference template of the unit and the marginal constraint value parameters of the dual space are subjected to discrete variable pre-locking processing to obtain the pre-locking result. The initial state consistency mandatory verification and correction processing is performed on the pre-locking result to obtain the locked and corrected discrete variable state. The locked and corrected discrete variable states and system physical constraints are coupled and pruned to obtain an initial simplified constraint set; The initial simplified constraint set and the system limit power flow distribution are subjected to constraint validity determination to obtain the target simplified constraint set; A power system dispatching original model with network constraints is constructed. The locked and corrected discrete variable states, the target simplified constraint set, and the power system dispatching original model with network constraints are compressed and solved to obtain the power system dispatching scheme.

2. The method according to claim 1, characterized in that, The process involves acquiring historical operating data and simulated forecast data of the power system, performing clustering and structural consistency assessment on the historical operating data and simulated forecast data, and obtaining a feature library of historical optimal dispatch patterns, including: The node net load is calculated and preprocessed based on the historical operating data and simulated prediction data to obtain a node net load data sample set; The node net load data sample set is subjected to unsupervised hierarchical partitioning to obtain multiple scene clusters; The structural consistency index is calculated for the optimal power system scheduling solution within each scenario cluster to obtain the scheduling scheme stability index corresponding to each scenario cluster. The stability index of the scheduling scheme is compared with the preset structural consistency threshold to obtain a feature library of historical optimal scheduling modes.

3. The method according to claim 1, characterized in that, The process of acquiring load and renewable energy scenario instances involves performing scenario matching processing between these instances and the historical optimal scheduling mode feature library to obtain the optimal unit scheduling reference template, including: The node net load feature is extracted from the load and new energy scenario examples to obtain the online net load feature vector; The similarity calculation is performed between the online net load feature vector and the central feature vector of each scenario cluster in the historical optimal scheduling mode feature library to obtain multiple scenario matching degree values. The multiple scene matching degree values ​​are sorted in descending order and extreme value filtering is performed to obtain the target scene cluster that is closest to the online instance; The intersection of stable scheduling schemes corresponding to the target scenario cluster is extracted and processed to obtain the optimal scheduling reference template for the unit.

4. The method according to claim 1, characterized in that, The process involves pre-locking discrete variables between the optimal scheduling reference template for the unit and the marginal constraint value parameters of the dual space to obtain a pre-locking result. Then, the pre-locking result undergoes initial state consistency mandatory verification and correction to obtain the locked and corrected discrete variable states, including: The load and new energy scenario examples are subjected to single-cycle dual relaxation solution processing to obtain the dual space marginal constraint value parameters. The dual space marginal constraint value parameters are processed by the unit state locking value calculation to obtain the unit state locking value index corresponding to each discrete variable. The unit status lock value index and the unit optimal scheduling reference template are sorted in descending order to obtain a discrete variable sequence. The discrete variable sequence and the preset reduction intensity coefficient are subjected to state locking processing to obtain the pre-locking result; The initial start-up and shutdown status, minimum continuous operating time, and minimum continuous stop time of the unit are obtained. The pre-locking result is subjected to constraint verification processing with the initial start-up and shutdown status, the minimum continuous operating time, and the minimum continuous stop time to obtain the constraint violation result. The pre-locked variables corresponding to the constraint violation results are corrected and locked to obtain the locked and corrected discrete variable states.

5. The method according to claim 1, characterized in that, The coupling constraint pruning process performed on the locked and corrected discrete variable states and system physical constraints yields an initial simplified constraint set, including: The locked and corrected discrete variable states are subjected to adjacent time period start-stop state matching processing to obtain the first target unit set; All ramping constraints corresponding to the first target unit set are removed to obtain the first stage simplified constraint set; The locked and corrected discrete variable states are subjected to minimum on / off time interval homomorphic locking identification processing with the first stage simplified constraint set to obtain the second target unit set. All minimum start-up and shutdown time constraints corresponding to the second target unit set are eliminated to obtain an initial simplified constraint set.

6. The method according to claim 1, characterized in that, The process of determining the constraint effectiveness of the initial simplified constraint set and the system limit power flow distribution to obtain the target simplified constraint set includes: The network transmission security constraints are processed into a compact matrix representation to obtain the network constraint matrix. The row elements corresponding to each network constraint in the network constraint matrix are rearranged in descending order to obtain the rearranged network constraint matrix. The upper limit of the physical output of the units corresponding to the elements of the rearranged network constraint matrix is ​​successively accumulated to obtain the critical unit index that satisfies the total net load boundary condition; The critical unit index and network constraint boundary are analyzed to construct and determine the boundary inequality, thereby obtaining a set of redundant network constraints. The initial simplified constraint set and the redundant network constraint set are subjected to constraint elimination processing to obtain the target simplified constraint set.

7. The method according to claim 1, characterized in that, The process involves constructing a power system dispatching primitive model with network constraints, and then performing a compressed solution process on the locked and corrected discrete variable states, the target simplified constraint set, and the power system dispatching primitive model with network constraints to obtain a power system dispatching scheme, including: The solution requirements for power system scheduling with network constraints are subjected to type matching processing to obtain the target solution type, which includes deterministic solution, multi-scenario stochastic solution and robust optimization solution; The solution type is compressed to adapt the locked and corrected discrete variable states, the target simplified constraint set, the target solution type, and the original power system dispatch model with network constraints to obtain a dimensionality-reduced solution model. The feasibility of the dimensionality reduction solution model is verified and iteratively solved to obtain the power system dispatching scheme.

8. A design device for a power grid dispatching scheme, characterized in that, include: The acquisition module is used to acquire historical operating data and simulated forecast data of the power system, perform clustering and structural consistency assessment on the historical operating data and simulated forecast data, and obtain a historical optimal dispatch mode feature library. The historical optimal dispatch mode feature library stores the implicit mapping relationship between node net load characteristics and unit optimal dispatch scheme. The acquisition module is also used to acquire load and new energy scenario instances, perform scenario matching processing on the load and new energy scenario instances and the historical optimal scheduling mode feature library to obtain the optimal scheduling reference template for the unit. The processing module is used to perform discrete variable pre-locking processing on the optimal scheduling reference template of the unit and the marginal constraint value parameters of the dual space to obtain the pre-locking result, and to perform initial state consistency mandatory verification and correction processing on the pre-locking result to obtain the locked and corrected discrete variable state. The processing module is also used to perform coupling constraint pruning on the locked and corrected discrete variable states and system physical constraints to obtain an initial simplified constraint set; The processing module is also used to perform constraint validity determination processing on the initial simplified constraint set and the system limit power flow distribution to obtain the target simplified constraint set; The construction module is used to construct the original power system dispatch model with network constraints, and to perform compression and solution processing on the locked and corrected discrete variable states, the target simplified constraint set, and the original power system dispatch model with network constraints to obtain the power system dispatch scheme.

9. A computer device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images