Regional power peak shaving resource allocation system based on cloud computing

Abstract

The application relates to the field of power system operation, and discloses a regional power peak regulation resource allocation system based on cloud computing, which is used for obtaining a parallel acceleration and engineering feasible peak regulation resource allocation scheme. The system comprises a resource processing module, a regional division module, an initial solution obtaining module, an iterative optimization module and an execution correction module. The resource processing module is used for constructing a discretized peak regulation resource pool; the regional division module is used for constructing a regional coupling graph based on the parameters of inter-regional tie lines, dividing the whole network into a plurality of dispatching sub-regions to accelerate internal priority search; the initial solution obtaining module is used for constructing a system total cost function and obtaining an initial feasible solution satisfying constraints; the iterative optimization module is used for obtaining candidate exchanges based on the dispatching sub-regions, performing constraint checking and calculating a cost drop, and then iteratively updating to obtain an allocation scheme; and the execution correction module is used for issuing the scheme to an edge node for execution and performing local correction according to feedback. The application improves the solution efficiency of large-scale peak regulation resources, reduces the total system dispatching cost, and guarantees the safe and stable operation of the power grid through cloud-edge cooperation.

Description

Technical Field

[0001] This invention relates to the field of power system operation, and more particularly to a regional power peak-shaving resource allocation system based on cloud computing. Background Technology

[0002] As the global energy structure transitions towards a low-carbon model, the proportion of new energy sources such as wind power and solar power in the power system continues to increase. However, the output of new energy sources is significantly volatile and intermittent due to weather conditions, leading to a widening peak-to-valley difference in the net load of the power grid (load minus new energy output), and increasingly severe pressure on system peak regulation. Modern power systems are typically composed of multiple regional power grids interconnected by tie lines, and different regions often exhibit temporal and spatial complementarity in load characteristics and new energy resources. Therefore, breaking down regional barriers and conducting large-scale, cross-regional collaborative optimization of peak-regulating resources has become a key approach to improving the absorption capacity of renewable energy and reducing the overall peak-regulating cost of the system.

[0003] Inter-regional peak-shaving resource allocation involves various peak-shaving resources, including thermal power, hydropower, energy storage, and interruptible loads. These resources include both continuously adjustable and discretely operating ones; their cost characteristics are often non-convex and non-linear. Meanwhile, regional power grids are vast in scale and subject to complex constraints. How to quickly and economically determine the allocation plan for peak-shaving resources in each region while ensuring safety is a key issue in the field of power system dispatching.

[0004] Currently, existing technologies for allocating regional power peak-shaving resources have the following main shortcomings: Traditional methods typically model both continuously adjustable and discrete resources in a hybrid manner, preserving non-convex cost curves, which transforms the problem into a large-scale mixed-integer nonlinear programming problem (MINLP). Solving such problems using commercial solvers is computationally extremely time-consuming, making it difficult to meet the time requirements of real-time scheduling. While heuristic methods such as genetic algorithms and particle swarm optimization offer faster solution speeds, they cannot guarantee the global optimality of the solution and the results are unstable.

[0005] Existing centralized or cloud-based optimization solutions typically employ a one-size-fits-all approach, with edge nodes merely acting as execution units, lacking local real-time security verification and autonomous decision-making capabilities. When solutions deployed from the cloud center cannot be executed due to local power grid security constraints (power flow exceeding limits), or when communication links experience delays or interruptions, edge nodes cannot effectively respond, resulting in insufficient overall system reliability and resilience.

[0006] Therefore, we propose a cloud-based regional power peak-shaving resource allocation system to address the aforementioned issues. Summary of the Invention

[0007] This invention provides a cloud computing-based regional power peak-shaving resource allocation system to obtain a peak-shaving resource allocation scheme that can be parallelized, accelerated, and implemented in engineering.

[0008] The first aspect of this invention provides a cloud computing-based regional power peak-shaving resource allocation system, comprising: a resource processing module, used to acquire peak-shaving resource information of each region, discretize the peak-shaving resources and handle non-convexity by using piecewise linear convex envoy relaxation to obtain discrete peak-shaving resource levels, and construct a discretized peak-shaving resource pool; a region division module, used to acquire parameters of the tie lines between the regions, construct a region coupling graph based on the tie line parameters, and divide the regions based on the region coupling graph to obtain multiple scheduling sub-regions; and an initial solution acquisition module, used to obtain the discrete peak-shaving resources in each time period. The system uses gear position as the decision variable to construct a total system cost function and obtains an initial feasible solution that satisfies the equipment operation constraints. An iterative optimization module is used to obtain candidate exchanges based on the scheduling sub-regions, perform constraint verification on the candidate exchanges, calculate the cost change corresponding to the verified candidate exchanges according to the total system cost function, and iteratively update the initial feasible solution based on the cost change to obtain the current allocation solution. An execution correction module is used to distribute the current allocation solution as a peak-shaving resource allocation scheme to the edge nodes of each region for execution, and perform local corrections to the peak-shaving resource allocation scheme based on the execution feedback information from the edge nodes.

[0009] Optionally, in a first implementation of the first aspect of the present invention, the resource processing module is specifically used for: The continuously adjustable resources in the peak-shaving resources are discretized into integer levels according to a preset step size; For the non-convex cost curve of the peak-shaving resource, a piecewise linear convex envelope relaxation model is constructed, and the non-convexity is handled in the objective function; Based on the processed peak-shaving resources, the corresponding integer levels, and the relaxed convex hull cost curve, a discrete peak-shaving resource pool is formed.

[0010] Optionally, in a second implementation of the first aspect of the present invention, the region partitioning module, when constructing a region coupling diagram based on the parameters of the connecting lines, is specifically used for: Obtain the equivalent reactance and transmission capacity of the tie line; The initial electrical coupling weight of the tie line is determined based on the equivalent reactance and the transmission capacity, and the initial electrical coupling weight is adjusted in combination with historical peak-shaving resource mutual assistance direction data to obtain the final coupling weight. A regional coupling graph is constructed using each region as a node and the connecting lines with the coupling weights as edges.

[0011] Optionally, in a third implementation of the first aspect of the present invention, when the region partitioning module partitions the regions based on the region coupling diagram to obtain multiple scheduling sub-regions, it is specifically used for: Calculate the Laplacian matrix corresponding to the region coupling graph; Find the eigenvector corresponding to the second smallest eigenvalue in the Laplacian matrix; The regions are recursively divided into two parts according to the positive or negative signs of the elements in the feature vector, until the number of regions contained in each sub-region after division meets the preset condition, thus obtaining multiple scheduling sub-regions.

[0012] Optionally, in a fourth implementation of the first aspect of the present invention, the total system cost function includes a resource call cost item, a cross-regional imbalance penalty item, a peak shaving insufficient penalty item, and a peak shaving excessive penalty item; The resource allocation cost item adopts a discrete convex function form, the cross-regional imbalance penalty item adopts a discrete convex function applied to the difference in the total peak-shaving power of different regions in each region, and the peak-shaving insufficient penalty item and the peak-shaving excess penalty item adopt discrete convex functions applied to the deviation between the net load and the total peak-shaving output of each region, respectively. The total cost function of the system satisfies the M-convex property over the integer feasible region of the decision variables.

[0013] Optionally, in a fifth implementation of the first aspect of the present invention, the formula for calculating the total system cost function is: ; In the formula: The total system scheduling cost; This refers to the discrete levels of peak-shaving resources at different times. This is a secondary cost coefficient. This is the primary cost coefficient; and These represent the total power already utilized by the two areas with interconnecting lines. For the coupling weights of the corresponding tie lines, This is the penalty coefficient for cross-regional imbalance; This is the power deviation between net load and total peak-shaving output. This is the deviation penalty coefficient.

[0014] Optionally, in a sixth implementation of the first aspect of the present invention, when the iterative optimization module obtains candidate swaps based on the scheduling sub-region, it is specifically used for: Prioritize obtaining the first candidate swap within any of the aforementioned scheduling sub-regions; If there is no first candidate exchange within the scheduling sub-region that reduces the total cost function of the system, then cross-regional exchange is used as a secondary backup strategy to obtain a second candidate exchange between different scheduling sub-regions. Both the first candidate exchange and the second candidate exchange are discrete gear shifts between the two peak-shaving resources.

[0015] Optionally, in the seventh implementation of the first aspect of the present invention, the device operation constraints and the constraint verification of the candidate exchange both include resource upper and lower limit constraints, ramp rate constraints, and tie line capacity constraints.

[0016] Optionally, in an eighth implementation of the first aspect of the present invention, a scene prediction module is further included, used for: Obtain net load prediction error data, and select multiple quantile levels based on the distribution of the net load prediction error data to form a typical net load scenario set containing scenario probabilities; The typical net load scenario set is used to calculate the expected cost for multiple scenarios when constructing the total cost function of the system.

[0017] Optionally, in a ninth implementation of the first aspect of the present invention, when the execution correction module performs local corrections to the peak-shaving resource allocation scheme based on the execution feedback information of the edge nodes, it is specifically used for: After receiving the edge node's local power flow security verification and execution of the peak shaving resource allocation scheme, the uploaded actual execution level and deviation reason are used as the execution feedback information. Based on the execution feedback information, the affected area is determined. While keeping the allocation scheme of other areas unchanged, a local candidate exchange is obtained within the affected area to update the peak-shaving resource allocation scheme.

[0018] The mechanism of this invention is as follows: By introducing multiple factors such as resource call cost, cross-regional imbalance penalty, and insufficient and excessive peak shaving penalties, it is designed as a discrete convex function with M-convex properties. A key property of M-convex functions is that local optima are also global optima. This ensures that the greedy iterative algorithm based on feasible descent exchanges can converge to the global optimal solution in a finite number of steps, avoiding the combinatorial explosion dilemma faced by traditional mixed integer programming in large-scale problems. The candidate exchanges are obtained by adopting a hierarchical search strategy that prioritizes exchanges within sub-regions and then exchanges between sub-regions, which greatly reduces the search space for each iteration, making it possible to compute the cost changes of each candidate exchange in parallel. This leverages the advantages of global optimization in the cloud center, utilizes the local real-time information of edge nodes to ensure security, and enhances the robustness of the scheme to prediction bias and actual execution error through a local correction mechanism. Beneficial effects: It abandons the traditional large-scale mixed integer programming or heuristic algorithm, discretizes continuous resources, linearizes non-convex costs into piecewise linear convex, and cleverly constructs a total cost function with M-convex properties, including cross-regional imbalance penalty and peak-shaving deviation penalty.

[0019] Beneficial effects: The M-convex property ensures that the local search algorithm based on resource slot exchange can monotonically decrease and converge to the global optimum, avoiding the defect of heuristic algorithms getting trapped in local optima. Since the decision variable is defined as the discrete slot of the peak-shaving resource, the slot exchange operation naturally satisfies the conservation condition of the sum of integer grid point components in the feasible region of the M-convex function. The cross-regional power imbalance is constrained by the penalty term in the objective function. Each iteration only needs to calculate the minimal incremental cost, which greatly reduces the algorithm complexity and can easily cope with the real-time optimization of massive peak-shaving resources. The partitioning results no longer depend on administrative boundaries, but match the actual electrical coupling strength of the power grid. Moreover, the partitioning direction is highly consistent with the historical mutual flow, which makes the candidate exchange strategy of prioritizing the search within the sub-region and then expanding to the interval highly targeted. It significantly prunes the invalid search space, gives full play to the parallel processing capabilities of cloud computing, and improves the iterative convergence speed by orders of magnitude. Local security verification at edge nodes compensates for local network constraints that may be overlooked in centralized modeling at the cloud center; and after receiving feedback, the cloud center does not need to perform global optimization again, but can eliminate deviations through lightweight local exchange corrections. This ensures both the economic optimality of the initial solution and the security and feasibility of actual execution, while also possessing strong fault tolerance. Without compromising the M-convex property, the optimization scheme is upgraded from deterministic single-point optimization to multi-scenario probabilistic optimization, which has strong adaptability and robustness to fluctuations in new energy output and load forecasting errors, and significantly reduces the penalty for insufficient or excessive peak shaving in actual operation. By limiting candidate exchanges to integer-level transfers between two resources within the same time period and combining this with regional power transfer distribution factors for rapid verification of tie-line capacity, the cost calculation and constraint verification of all candidate exchanges are completely independent and naturally parallel, thus maximizing the utilization of distributed computing resources in cloud computing. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of an embodiment of a cloud computing-based regional power peak-shaving resource allocation system according to the present invention. Figure 2 This is a schematic diagram of another embodiment of the regional power peak-shaving resource allocation system based on cloud computing in this invention. Figure 3 This is a schematic diagram of an embodiment of a cloud-based regional power peak-shaving resource allocation device according to the present invention; Figure 4 The diagram shows the process of transforming the original non-convex cost curve into a piecewise linear convex curve and discretizing continuous resources into integer increments. The left side shows the original curve with decreasing marginal cost, and the right side shows the convex curve with increasing marginal cost after the correction. Detailed Implementation

[0021] This invention provides a cloud-based regional power peak-shaving resource allocation system for obtaining a peak-shaving resource allocation scheme that can be parallelized, accelerated, and implemented in engineering. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the regional power peak-shaving resource allocation system based on cloud computing in this invention includes: 101. Resource processing module, used to collect peak-shaving resource information in various regions, discretize continuously adjustable resources into integer levels with a fixed step size, directly represent discrete level resources as integer levels, and perform piecewise linear convexification processing on non-convex cost curves to form a discretized peak-shaving resource pool.

[0023] It is understood that the executing entity of this invention can be a regional power peak-shaving resource allocation device based on cloud computing, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0024] It should be noted that the cloud center has currently collected information on three typical peak-shaving resources within its jurisdiction: The gas turbine unit located at node A (industrial park) has a capacity reduction range of 0-30MW. The system is set with a step size of 10MW, discretized into four integer levels: level 0 (0MW), level 1 (10MW), level 2 (20MW), and level 3 (30MW). The reported peak-shaving costs for levels 1-3 are 280 yuan, 500 yuan, and 800 yuan, respectively. The system found that the marginal cost increment for level 0-1 is 280 yuan, while the increment for level 1-2 is only 220 yuan, exhibiting non-convex characteristics. This solution uses piecewise linear convex envelope relaxation to handle the non-convexity. After processing, mathematically, the cost curve used by this relaxation model (e.g., level 1 adjustment to 240 yuan, level 2 500 yuan) is the lower bound of the original non-convex problem, successfully transforming it into a convex cost curve.

[0025] The electric arc furnace located at node A has two independent power lines that can be interrupted at any time, with each line having a fixed power of 10MW. The system represents this as three integer levels, with levels 0 to 2 corresponding to the interruption of 0 to 2 lines, respectively. The total compensation cost for each level is 0 yuan, 400 yuan, and 850 yuan, respectively, and it has an inherent convex property.

[0026] The energy storage power station located at node C (High-tech Industrial Park) has a usable discharge capacity of 10MW. The system is discretized into three integer levels in 5MW increments: level 0 (0MW), level 1 (5MW), and level 2 (10MW). The costs for each level are 0 yuan, 150 yuan, and 350 yuan, respectively, and the marginal cost satisfies the convexity requirement. After the above processing, the system integrates the above resources and inputs them into the standard discretized peak-shaving resource pool.

[0027] 102. The region partitioning module is used to construct a region coupling graph based on the electrical parameters of the inter-regional tie lines, solve for the Fiedler vector corresponding to the second smallest eigenvalue of the Laplace matrix of the region coupling graph, and recursively bisect the region according to the sign of the Fiedler vector to form multiple scheduling sub-regions.

[0028] It should be noted that, in addition to managing nodes A (industrial park) and C (high-tech industrial park), the cloud center also manages nodes B (large commercial area) and D (regional energy storage hub). These four nodes are connected in pairs to form a regional network.

[0029] Cloud center data collection connection capacity weights: The capacity between node A and node C is relatively large at 50MW; the capacity between node B and node D is also 50MW. However, the connection between node A and node B, and between node C and node D, is relatively weak, both at 10MW.

[0030] The cloud center constructs a graph model and derives the Laplacian matrix. After running the eigenvalue decomposition algorithm, the Fiedler vector is obtained. The values ​​corresponding to the four nodes show differentiation: node A is a positive number 0.45, node C is a positive number 0.52; node B is a negative number -0.48, and node D is a negative number -0.55.

[0031] The system divides nodes into scheduling sub-regions based on signs, merging nodes A (industrial park) and C (high-tech industrial park) with both positive signs into scheduling sub-region 1; and merging nodes B and D with both negative signs into scheduling sub-region 2. This partition is only used for generating the initial solution and internal priority search; the physical constraints of cross-region connections are still retained in subsequent global optimization.

[0032] 103. The initial solution acquisition module is used to construct the total system cost function with M-convex properties using the discrete levels of each resource in each time period as decision variables, and to obtain the initial feasible solution that satisfies the upper and lower limit constraints of resources and the ramp constraint.

[0033] It should be noted that, for scheduling sub-region 1, the system formulates an initial plan for time period 1. The resource pool is known as follows: Node A has a gas turbine (maximum 3 levels, with a lower limit of cost of 240 yuan for level 1 and 500 yuan for level 2) and an electric arc furnace (maximum 2 levels, with a cost of 400 yuan for level 1); Node C has energy storage (maximum 2 levels, with a cost of 150 yuan for level 1 and 350 yuan for level 2).

[0034] Assume node A has a demand of 20MW and node C has a demand of 7MW. The system allocates node A's gas turbine to level 2 (20MW, allocation cost 500 yuan), node C's energy storage to level 1 (5MW, allocation cost 150 yuan), and node A's electric arc furnace to level 0. The direct resource allocation cost is 650 yuan. Adding the cross-regional imbalance penalty of 300 yuan and the peak-shaving deviation penalty of 40 yuan, the total system scheduling cost of the initial feasible solution is 990 yuan. This solution satisfies local peak-shaving and physical limits, reserving room for cost reduction in cross-node resource exchange.

[0035] 104. Iterative optimization module, used to obtain candidate exchanges within and between scheduling sub-regions, verify the resource upper and lower limits, ramp rate and tie line capacity constraints of candidate exchanges, calculate the change in total system cost corresponding to each candidate exchange in parallel, select the candidate exchange with the largest cost reduction and pass the verification to update the current solution, and repeat the iteration until there are no feasible exchanges to reduce the cost.

[0036] It should be noted that, taking time period 1 of sub-region 1 as an example, the initial solution is as follows: gas turbine at node A, level 2 (20MW, 500 yuan), energy storage at node C, level 1 (5MW, 150 yuan), electric arc furnace at node A, level 0, with a total cost of 990 yuan. The remaining capacity of the connecting line between nodes A and C is 40MW.

[0037] The system obtains and verifies cross-node candidate switching schemes, as detailed in Table 1 below: Table 1

[0038] After parallel comparison, the system found that the total cost change of Scheme 1 was positive (+690 yuan) and was therefore eliminated. Due to the asymmetry of the step size of different resources, a simple swap directly disrupted the original power balance, leading to a significant increase in the penalty term. After scanning the same sub-region and cross-regional backup strategies, no feasible swap schemes to further reduce costs were found. The system determined that the current solution (node ​​A gas turbine at step 2, node C energy storage at step 1, electric arc furnace at step 0, total cost 990 yuan) met the optimal conditions, and the iterative convergence was achieved and output.

[0039] 105. The execution correction module is used to distribute the iteratively updated current solution as a peak-shaving resource allocation scheme to the edge nodes of each region. The edge nodes perform local security verification and then execute the scheme, and feed back the actual execution status to the cloud center. The cloud center performs local exchange correction on the affected areas based on the feedback, and obtains the corrected peak-shaving resource allocation scheme.

[0040] It should be noted that the cloud center issued the optimal solution for time period 1: requiring the gas turbine of node A to be controlled at level 2 (20MW), requiring the energy storage of node C to be controlled at level 1 (5MW), and keeping the electric arc furnace of node A at level 0.

[0041] During the execution phase, the local verification of the edge node at node C passed, and the energy storage power station successfully output 5MW. When the edge node at node A controlled the gas turbine to upgrade to level 2, it detected a restriction in the intake mechanical valve. For safety reasons, the local verification of the upgrade failed, and the gas turbine was forcibly locked at level 1 (10MW), and the cause of the deviation was reported to the cloud.

[0042] The cloud center detected a 10MW peak-shaving shortfall at Node A during the specified time period (actual output was 10MW, demand was 20MW, and a high deviation penalty would be imposed). The cloud center immediately initiated a local correction within sub-region 1. A scan revealed that Node A's electric arc furnace was at level 0, ready to respond to the 10MW command at any time (compensation cost 400 yuan), and there was no risk of cross-node interconnection power flow exceeding limits. The cloud center then issued a correction command to Node A. The edge node of Node A passed the second verification and the 10MW electric arc furnace load was removed. The correction plan was changed to: Node A gas turbine at level 1, Node A electric arc furnace at level 1, and Node C energy storage at level 1. The system closed-loop completed the peak-shaving task and filled the shortfall.

[0043] Please see Figure 2 , Figure 4 Another embodiment of the regional power peak-shaving resource allocation system based on cloud computing in this invention includes: 201. Resource processing module, used to collect peak-shaving resource information in various regions, discretize continuously adjustable resources into integer levels with a fixed step size, directly represent discrete level resources as integer levels, and perform piecewise linear convexization processing on non-convex cost curves to form a discretized peak-shaving resource pool.

[0044] Specifically, the system collects the resource type, maximum technical output, minimum technical output, cost price, ramp rate, and energy storage state of charge of peak-shaving resources in each region; continuously adjustable resources are discretized into integer levels according to a preset step size, and the original level value is retained for discrete level resources; non-convex cost curves are made convex using a piecewise linear approximation method, so that the convex cost curve is linear in each segment and the overall slope increases; the resulting discretized peak-shaving resource pool includes the range of integer levels available for each resource in each time period, the cost function after discretization and convexization, and the upper and lower limits of the ramp rate.

[0045] It should be noted that the cloud center is currently processing peak-shaving resources in an industrial area. It collects multi-dimensional information on three typical resource types through the gateway and performs the following processing: Continuously adjustable resources: Node A acquires data from one gas turbine with a maximum technical output of 30MW and a minimum technical output of 0MW. The ramp rate is limited to a maximum of one ramp per scheduling period. The system sets a fixed step size of 10MW, discretizing it into four integer ramps: ramp 0 (0MW), ramp 1 (10MW), ramp 2 (20MW), and ramp 3 (30MW). The reported peak-shaving total cost quotes from the unit for each ramp are 0 yuan, 280 yuan, 500 yuan, and 800 yuan, respectively. The system calculates the marginal cost and finds that the increment from ramp 0 to ramp 1 is 280 yuan, and the increment from ramp 1 to ramp 2 is 220 yuan, indicating a decrease in marginal cost. The system uses piecewise linear convex enveloping relaxation to handle non-convexity. After relaxation, in the mathematical model input to the optimization solver, the lower bound of the gas turbine cost for level 1 is mapped to 240 yuan, and for level 2 it is 500 yuan, forming a convex envelope curve with marginally increasing costs (240 yuan and 260 yuan), thus being included in the standard discretized peak-shaving resource pool.

[0046] Secondly, there are discrete power level resources. Node A has two electric arc furnaces that can be interrupted at any time, with a maximum technical output of 20MW and a minimum technical output of 0MW, allowing each time period to span two power levels. The system directly maps these to three integer power levels: Power Level 0 (0MW), Power Level 1 (10MW), and Power Level 2 (20MW). The total compensation cost quotes for each power level are 0 yuan, 400 yuan, and 850 yuan, respectively. The calculated marginal costs are 400 yuan and 450 yuan, respectively, exhibiting inherent convexity, and their original cost values ​​are retained unchanged.

[0047] Energy storage resources: Node C has one energy storage power station with a maximum technical output of 10MW and a minimum technical output of 0MW. The current state of charge (SOC) of the energy storage is 65%, allowing for two shifts per time period. The system discretizes it into three integer shifts in 5MW increments: Shift 0 (0MW), Shift 1 (5MW), and Shift 2 (10MW). The peak-shaving costs for each shift are 0 yuan, 150 yuan, and 350 yuan, respectively, with marginal costs of 150 yuan and 200 yuan, satisfying the convexity requirement. After the above processing, the system unifies and structures the availability range of the above resources, the corrected convexity cost, and the upper and lower limit ramp rates, storing them in the cloud system to form a complete discrete peak-shaving resource pool for the current region.

[0048] 202. The region partitioning module is used to construct a region coupling graph based on the electrical parameters of the inter-regional tie lines, solve for the Fiedler vector corresponding to the second smallest eigenvalue of the Laplace matrix of the region coupling graph, and recursively bisect the region according to the sign of the Fiedler vector to form multiple scheduling sub-regions.

[0049] Specifically, the coupling weights between regions are determined based on the equivalent reactance and transmission capacity of each interconnecting line between regions, and a region coupling graph is constructed with regions as nodes and interconnecting lines as edges; the Laplacian matrix of the region coupling graph is calculated, and the Fiedler vector corresponding to the second smallest eigenvalue of the Laplacian matrix is ​​solved; the region set is divided into two sub-regions according to the positive or negative sign of the corresponding elements of each region in the Fiedler vector; the region coupling graph construction and Fiedler vector division are repeated for each sub-region until the number of regions contained in each sub-region does not exceed a preset upper limit; The multiple scheduling sub-regions formed are electrically tightly coupled within each other, and electrically weakly coupled between each other. The multiple scheduling sub-regions are used to limit the priority search range when candidate exchanges are obtained, that is, candidate exchanges are obtained first within the scheduling sub-regions, and candidate exchanges are obtained between the scheduling sub-regions only when there are no feasible descending exchanges within the scheduling sub-regions.

[0050] Furthermore, based on the equivalent reactance and transmission capacity of each interconnecting line between regions, the coupling weight between regions is determined, and a regional coupling graph is constructed with regions as nodes and interconnecting lines as edges. This includes: obtaining the equivalent reactance value, transmission capacity value, and historical peak-shaving resource mutual assistance direction data of each interconnecting line between regions; calculating the initial electrical coupling weight based on the equivalent reactance value of each interconnecting line; determining the upper limit cutoff value of the weight based on the transmission capacity value; and truncating the initial electrical coupling weight. The consistency coefficient of peak-shaving mutual assistance direction between regions is determined based on historical peak-shaving resource mutual assistance direction data. A larger value is taken when peak-shaving resources have historically flowed mainly from one region to another, and a smaller value is taken when bidirectional mutual assistance or historical mutual assistance is less frequent. The truncated initial electrical coupling weight is multiplied by the consistency coefficient of peak-shaving mutual assistance direction to obtain the coupling weight corresponding to each tie line. A regional coupling graph is constructed with regions as nodes and tie lines with coupling weights as edges. The regional coupling graph is used to calculate the Laplace matrix and solve the Fiedler vector to ensure that the partitioning results are consistent with the historical peak-shaving mutual assistance direction.

[0051] It should be noted that the power grid managed by the cloud center comprises four distribution areas: node A, node B, node C, and node D. The maximum size of the dispatch sub-area is set to two nodes. The coupling weights of the interconnections between the four nodes are now calculated: For the tie line AC between node A and node C, the initial electrical coupling weight is measured to be 60. The physical transmission capacity limit is 50MW, and the initial weight is truncated to 50. Historical data shows that node A frequently transmits peak-shaving resources unidirectionally to node C, with high consistency in the direction of peak-shaving mutual assistance; the consistency coefficient is set to 1.2. The coupling weight is 50 multiplied by 1.2, which equals 60.

[0052] For the tie line BD between node B and node D, the initial electrical coupling weight is 55, and the maximum transmission capacity is 60MW, retaining the initial value of 55. Historical data shows that node D has long provided peak-shaving support to node B in a unidirectional manner, and the consistency coefficient is set to 1.2. The coupling weight is 55 multiplied by 1.2, which equals 66.

[0053] For the tie line AB between node A and node B, the initial electrical coupling weight is 15, the transmission capacity limit is 20MW, and 15 is reserved. Historical data shows that peak-shaving resources between the two locations frequently flow bidirectionally, so the consistency coefficient is set to 0.8. The coupling weight is 15 multiplied by 0.8, which equals 12.

[0054] For the tie line CD between node C and node D, the initial electrical coupling weight is 12, the maximum transmission capacity is 10MW, and the truncation weight is 10. Historical data shows that the two locations rarely support each other, so the consistency coefficient is set to 0.8. The coupling weight is 10 multiplied by 0.8, which equals 8.

[0055] Based on nodes A, B, C, and D and the weights of the four edges mentioned above, the cloud center constructs a regional coupling graph, derives the Laplacian matrix, and solves for the Fiedler vector. The calculations show that the numerical distribution of each node is as follows: node A is +0.42, node C is +0.46; node B is -0.45, and node D is -0.51.

[0056] The system segments nodes by sign, merging nodes A and C (both positive) into scheduling sub-region 1; and merging nodes B and D (both negative) into scheduling sub-region 2. Each sub-region has two nodes, which does not exceed the preset limit, so the current clustering terminates. This partition is only used for generating initial solutions and internal priority searches to accelerate convergence. Global optimization, used as a secondary backup strategy in subsequent cross-region exchanges, must ultimately retain and consider the physical constraints of all cross-region connections.

[0057] 203. Initial solution acquisition module is used to construct the total system cost function with M-convex properties using the discrete levels of each resource in each time period as decision variables, and to obtain the initial feasible solution that satisfies the upper and lower limit constraints of resources and the ramp constraint.

[0058] Specifically, the discrete levels of each resource in each time period are used as decision variables; a total system cost function is constructed, which consists of a resource allocation cost term, a cross-regional imbalance penalty term, a peak-shaving deficiency penalty term, and a peak-shaving excess penalty term; the resource allocation cost term adopts a discrete convex function form, the cross-regional imbalance penalty term adopts a discrete convex function applied to the difference in total peak-shaving power in different regions, and the peak-shaving deficiency penalty term and the peak-shaving excess penalty term adopt discrete convex functions applied to the deviation between the regional net load and the total peak-shaving output, respectively; the total system cost function satisfies the M-convex property on the integer feasible region of the decision variables; When an initial feasible solution is obtained, the peak-shaving resource levels of each region are called in order of increasing resource unit adjustment cost until the peak-shaving demand of each time period is met or the resources reach the available upper limit. During the allocation process, the upper and lower limit constraints of resources and the ramping constraints of adjacent time periods are verified to obtain an initial feasible solution that satisfies the upper and lower limit constraints of resources and the ramping constraints. The constructed total system cost function is used to calculate the cost change corresponding to each candidate exchange, and the initial feasible solution serves as the starting point for iterative updates.

[0059] Furthermore, the resource allocation cost term in the system's total cost function adopts a quadratic discrete convex function, where the cost of each resource in each time period is a quadratic function of the resource's discrete level; the cross-regional imbalance penalty term adopts an absolute value discrete convex function, where for each pair of regions with a tie line, the absolute value of the difference between the total peak-shaving levels of the two regions is taken, and then multiplied by the coupling weight of the tie line; the insufficient peak-shaving penalty term and the excessive peak-shaving penalty term each adopt a piecewise quadratic discrete convex function, where the total peak-shaving output of a region is lower than the net load, and the total peak-shaving output of a region is higher than the net load, also based on the square of the deviation; the system's total cost function is used to calculate the cost change caused by each candidate exchange.

[0060] It should be noted that the core task is to transform the discretized peak-shaving resources into computable decision variables, construct a system total cost evaluation framework that comprehensively considers economic efficiency and grid physical balance, and obtain an initial feasible solution. The specific implementation includes the following steps: 1. Establish decision variables: Use the integer position of peak-shaving resources in each scheduling period as decision variables.

[0061] 2. Constructing the total cost function: Taking into account resource allocation costs, inter-regional imbalance penalties, and peak-shaving deviation penalties, a system total cost function with M-convex properties is constructed. This function not only evaluates the direct costs of the allocated equipment but also constrains the differences in peak-shaving levels between regions and the deviation between peak-shaving supply and demand power.

[0062] 3. Obtaining the initial feasible solution: The system calculates the unit adjustment marginal cost for each resource level. Peak-shaving demand is allocated to the resources ranked highest according to cost from lowest to highest, with real-time verification of resource upper and lower limits and ramp-up rates until the regional peak-shaving demand is met.

[0063] Here, the following formula for calculating the total system cost is constructed in this scheme:

[0064] Formula parameter description: The total system scheduling cost is expressed in yuan.

[0065] In the first item, For discrete gears, the dimension is (gear); This is a secondary cost coefficient, with dimensions (yuan / grade). 2 ); This is a primary cost coefficient, measured in yuan per unit. The total unit offset for this item is yuan.

[0066] In the second item, and The total power already mobilized in the region is expressed in MW. The dimensionless coupling weights of the connection lines; The cross-regional imbalance penalty coefficient is expressed in yuan / MW. The overall dimension of this item is yuan.

[0067] In the third item, The deviation power is expressed in units of (MW). The deviation penalty coefficient is expressed in units of (yuan / MW). 2 The overall dimension of this item is (yuan).

[0068] Cross-region imbalance penalty coefficient setting: =1 / 3 yuan / MW. Cross-regional imbalance penalty: Node A total power 20MW, Node C total power 5MW, absolute power difference 15MW. Amount is 1 / 3. 60 15 = 300 yuan.

[0069] Peak shaving deviation penalty: Node A outputs 20MW, demand is 20MW, no surplus (0MW); Node C outputs 5MW, demand is 7MW, insufficient -2MW. The amount is 10×(0). 2 +10×(-2) 2 =40 yuan.

[0070] Initial total cost: 650 + 300 + 40 = 990 yuan. This solution serves as the input for the next iteration.

[0071] 204. Iterative optimization module, used to obtain candidate exchanges within and between scheduling sub-regions, verify the resource upper and lower limits, ramp rate and tie line capacity constraints of candidate exchanges, calculate the change in total system cost corresponding to each candidate exchange in parallel, select the candidate exchange with the largest cost reduction and pass the verification to update the current solution, and repeat the iteration until there are no feasible exchanges to reduce the cost.

[0072] Specifically, using the initial feasible solution as the current solution, a first candidate exchange set is obtained within the scheduling sub-region. Each first candidate exchange is a shift between two resources within the same scheduling sub-region. For each first candidate exchange in the first candidate exchange set, resource upper and lower limits, ramp rate, and tie-line capacity are verified. The first candidate exchanges that pass the verification constitute the first feasible exchange set. The change in total system cost corresponding to each first candidate exchange in the first feasible exchange set is calculated in parallel. If there is a first candidate exchange with a negative cost change, the first candidate exchange with the largest cost reduction is selected to update the current solution, and the process is repeated to obtain the first candidate exchange set. If there is no first candidate exchange with a negative cost change in the first feasible exchange set, a second candidate exchange set is obtained between scheduling sub-regions. Each second candidate exchange is a shift between two resources in different scheduling sub-regions. For each second candidate exchange in the second candidate exchange set, resource upper and lower limits, ramp rate, and tie-line capacity are checked. The second candidate exchanges that pass the checks form the second feasible exchange set. The total system cost change corresponding to each second candidate exchange in the second feasible exchange set is calculated in parallel. If there is a second candidate exchange with a negative cost change, the second candidate exchange with the largest cost reduction is selected to update the current solution, and the process returns to obtain the first candidate exchange set. If there is no second candidate exchange with a negative cost change in the second feasible exchange set, the current solution is determined to satisfy the M-convex function optimality condition, and the current solution is output as the peak-shaving resource allocation scheme.

[0073] Furthermore, the candidate switching is limited to reducing the discrete level of the first resource in the first region by an integer switching amount and increasing the discrete level of the second resource in the second region by an integer switching amount within the same scheduling period. The ramp rate verification includes: obtaining the discrete level of the first resource and the second resource in the adjacent scheduling periods of the scheduling period, calculating the difference between the discrete level of the scheduling period after the switching and the discrete level of the adjacent scheduling periods, and determining whether the difference is between the upper limit and the lower limit of the ramp rate of the corresponding resource. The tie line capacity verification includes: calculating the power change of each tie line caused by the candidate switching according to the pre-calculated regional power transfer distribution factor matrix, adding the power change to the current power flow value of each tie line, and determining whether the sum does not exceed the upper limit of the transmission capacity of the corresponding tie line.

[0074] It should be noted that the current initial solution is: Node A gas turbine in sub-region 1 is at level 2 (20MW), Node C energy storage power station is at level 1 (5MW), and Node A electric arc furnace is at level 0, with a total cost of 990 yuan.

[0075] The system obtains the first candidate exchange set within sub-region 1. Taking exchange scheme A as an example: reduce the gas turbine at node A by one level (down to level 1, a reduction of 10MW), and increase the energy storage at node C by one level (up to level 2, an increase of 5MW).

[0076] Regarding constraint verification, the energy storage was upgraded to level 2 without exceeding the upper limit, meeting the ramp-up rate requirement. The PTDF matrix calculated a power flow change of 5MW in the tie-line AC, which, after superposition, was below the 50MW upper limit, thus passing the physical verification.

[0077] In terms of cost change calculation, the resource allocation cost is reduced by 260 yuan due to the gas turbine downgrading (from 500 yuan to 240 yuan after convex enveloping relaxation), while the energy storage upgrade increases the cost by 200 yuan, resulting in a net reduction of 60 yuan. Regarding the inter-regional imbalance penalty, the output of node A is reduced to 10MW, while the output of node C is increased to 10MW, making the absolute value of the power difference 0MW, thus reducing this cost from 300 yuan to 0 yuan. However, in terms of peak-shaving deviation penalty, node A experiences a -10MW under-shaving deviation (local penalty increased to 1000 yuan), and node C experiences a +3MW over-shaving deviation (local penalty increased to 90 yuan). The overall net reduction in resource allocation cost is 60 yuan, and the total cost change of this scheme is calculated to be +690 yuan (i.e., -60 - 300 + 1090 - 40). Due to the increased net cost and the direct disruption of the original power balance, this scheme is accurately eliminated.

[0078] Because of the asymmetry in the step size of different resources (10MW and 5MW), a simple swap of the steps will directly disrupt the original power balance, resulting in a significant increase in the deviation penalty term. This allows the optimization algorithm to accurately identify and eliminate the resource, which proves the robustness of the total cost function in maintaining supply and demand balance.

[0079] The system subsequently obtained a second candidate exchange set across sub-regions. Calculations showed that the cross-regional power transfer output also could not compensate for the penalty cost caused by local peak-shaving deviations. Since no exchange scheme with negative cost change was found, the system determined that the initial feasible solution met the optimal conditions and output it as the peak-shaving resource allocation scheme.

[0080] 205. The execution correction module is used to distribute the iteratively updated current solution as a peak-shaving resource allocation scheme to the edge nodes of each region. The edge nodes perform local security verification and then execute the scheme, and feed back the actual execution status to the cloud center. The cloud center performs local exchange correction on the affected areas based on the feedback, and obtains the corrected peak-shaving resource allocation scheme.

[0081] Specifically, the cloud center packages the peak-shaving resource allocation plan by region and distributes it to the edge nodes of each region through an encrypted communication channel. After receiving the peak-shaving resource allocation plan, each edge node performs power flow safety verification based on local real-time measurement data. If the verification passes, it performs peak-shaving resource adjustment according to the peak-shaving resource allocation plan. If the verification fails, it stops execution and reports the verification failure information to the cloud center. After execution, each edge node uploads the actual execution level, response time, and reason for deviation to the cloud center. The cloud center determines the affected areas based on the actual execution levels and reasons for deviations reported by edge nodes in each region. While keeping the allocation schemes of other regions unchanged, it obtains local candidate exchanges within the affected areas and performs local exchange corrections to obtain the corrected peak-shaving resource allocation scheme. If the edge nodes do not receive the peak-shaving resource allocation scheme within the preset timeout period, they switch to local security mode and perform autonomous peak-shaving adjustment based on the most recently successfully received peak-shaving resource allocation scheme and local real-time measurement data.

[0082] It should be noted that the optimal solution for time period 1 is: gas turbine at node A at level 2 (20MW), electric arc furnace at node A at level 0 (0MW), and energy storage power station at node C at level 1 (5MW). The solution distribution and feedback are shown in Table 2 below: Table 2

[0083] During the execution phase, the energy storage power station at Node C successfully output 5MW. When the edge node at Node A controlled the gas turbine to upgrade to level 2, it detected a restriction in the intake mechanical valve. The local node determined that the upgrade verification failed, forcibly locking the gas turbine at level 1 (10MW), and reported the cause of the deviation to the cloud.

[0084] The cloud center discovered a 10MW peak-shaving shortfall at node A. If left uncorrected, node A's actual output is 10MW, while the corresponding demand is 20MW, resulting in a 10×(-10) MW peak-shaving deficit. 2 =A deviation penalty of 1000 yuan.

[0085] To fill the gap, the cloud center initiated a local correction within node A. A scan revealed that the electric arc furnace was at level 0, with level 1 having a step size of 10MW, and the incremental cost of calling it was 400 yuan. This incremental cost was lower than the 1000 yuan penalty for not correcting, making it economically feasible. After the command was issued, node A removed 10MW of electric arc furnace load, restoring the total output to 20MW and filling the gap.

[0086] Meanwhile, due to the communication interruption, node D automatically entered local security mode after a 5-second timeout, maintaining operation at speed 0 and ensuring system security during network outages.

[0087] 206. The scenario prediction module is used to acquire net load prediction data and renewable energy output prediction data for future scheduling periods in each region. Based on the net load prediction error distribution, multiple quantile levels are selected to form a typical net load scenario set. Each typical net load scenario includes the net load value for each region and time period. When constructing the total system cost function, the module calculates the peak-shaving deficiency penalty term and peak-shaving excess penalty term based on the typical net load scenario set. For each typical net load scenario, the module calculates the peak-shaving deficiency deviation and peak-shaving excess deviation for each region and sums them by scenario probability weighting. When calculating the total system cost change corresponding to each candidate exchange, the module calculates the expected cost change for multiple scenarios based on the typical net load scenario set. When verifying tie-line capacity, the maximum value of tie-line power change under all typical net load scenarios is used as the verification basis.

[0088] Furthermore, based on the net load prediction error distribution, multiple quantile levels are selected to form a typical net load scenario set, including: obtaining historical net load prediction error samples, statistically analyzing the net load prediction error for each scheduling period in each region, and fitting the prediction error probability distribution function corresponding to each scheduling period; selecting multiple preset quantile levels, and for each scheduling period, calculating the prediction error value at each quantile level based on the corresponding prediction error probability distribution function. The net load forecast values ​​for each region and time period are added to the forecast error values ​​at each quantile level to obtain the typical net load scenario corresponding to each quantile level. Each typical net load scenario includes the net load values ​​for each region and time period. The scenario probability of each typical net load scenario is determined according to the probability interval width between adjacent quantile levels. The sum of the scenario probabilities of all typical net load scenarios is 1. The typical net load scenario set and scenario probability are used to calculate the penalty for insufficient peak shaving and the penalty for excessive peak shaving, and are also used to calculate the expected cost change of multiple scenarios.

[0089] It should be noted that the cloud center needs to handle the uncertainty of node A's allocation in time period 2. Node A is significantly affected by the fluctuations of distributed photovoltaic power.

[0090] The basic net load forecast for period 2 is 25MW. Historical samples fitted by the cloud center show that the error follows a correlation distribution. The system selects three typical quantile levels to obtain the scenario and performs evaluation. When evaluating the scheme of providing 25MW of local peak-shaving output at node A, the deviation penalty coefficient is set to 10, and the calculated data is shown in Table 3 below: Table 3

[0091] According to Table 3, the system calculates the deviation penalty using a probability-weighted summation. The low-scenario penalty of 160 yuan is multiplied by a probability of 0.2, plus the baseline scenario penalty of 0 yuan multiplied by a probability of 0.6, and then the high-scenario penalty of 160 yuan multiplied by a probability of 0.2. The calculated expected deviation penalty cost, considering uncertainty, is 64 yuan. This value will be included in the total expected cost of the solution for subsequent optimization.

[0092] During the cross-regional tie-line capacity verification, the system directly retrieves extreme power flow variation values ​​under a high net load scenario with demand reaching 29MW. The scheme is only approved if the tie-line power flow does not exceed the physical transmission limit after the superposition of extreme power fluctuations, ensuring a safety margin in the execution of peak-shaving and allocation commands in the physical power grid.

[0093] Figure 3 This is a schematic diagram of a cloud-based regional power peak-shaving resource allocation device 300 provided in an embodiment of the present invention. The cloud-based regional power peak-shaving resource allocation device 300 can vary considerably due to differences in configuration or performance. The device 300 includes a transmitter 301, a receiver 302, and a processor 303. The processor 303 can also be a controller. Figure 3 The device is designated as "controller / processor 303". Optionally, the device 300 may also include a modem processor 305, which may include an encoder 306, a modulator 307, a decoder 308, and a demodulator 309.

[0094] In one example, transmitter 301 modulates (e.g., analog-to-analog conversion, filtering, amplification, and up-conversion, etc.) the output sample to obtain an uplink signal, which is transmitted via an antenna to an access network device. On the downlink, the antenna receives the downlink signal transmitted by the access network device. Receiver 302 modulates (e.g., filtering, amplification, down-conversion, and digitization, etc.) the signal received from the antenna and provides an input sample. In modem processor 305, encoder 306 receives service data and signaling messages to be transmitted on the uplink and processes (e.g., formatting, encoding, and interleaving) the service data and signaling messages. Modulator 307 further processes (e.g., symbol mapping and modulation) the encoded service data and signaling messages and provides an output sample. Demodulator 309 processes (e.g., demodulates) the input sample and provides a symbol estimate. Decoder 308 processes (e.g., deinterleaving and decoding) the symbol estimate and provides decoded data and signaling messages to device 300. Encoder 306, modulator 307, demodulator 309, and decoder 308 can be implemented by a combined modem processor 305. These units perform processing according to the radio access technology adopted by the radio access network (e.g., LTE and other evolved systems access technologies). It should be noted that when device 300 does not include modem processor 305, the above-mentioned functions of modem processor 305 can also be performed by processor 303.

[0095] The processor 303 controls and manages the operation of the device 300, and is used to execute the processing procedures performed by the device 300 in the above embodiments of this disclosure. For example, the processor 303 is also used to execute various steps of the transmitting or receiving device in the above method embodiments, and / or other steps of the technical solutions described in the embodiments of this disclosure.

[0096] Furthermore, the device 300 may also include a memory 304 for storing program code and data for the device 300.

[0097] Understandable Figure 3 Only a simplified design of device 300 is shown. In practical applications, device 300 can include any number of transmitters, receivers, processors, modem processors, memory, etc., and all devices that can implement the embodiments of this disclosure are within the protection scope of the embodiments of this disclosure.

[0098] The present invention also provides a cloud computing-based regional power peak shaving resource allocation device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor performs the steps of the cloud computing-based regional power peak shaving resource allocation system in the above embodiments.

[0099] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the cloud-based regional power peak-shaving resource allocation system.

[0100] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0101] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0102] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A regional power peak-shaving resource allocation system based on cloud computing, characterized in that, include: The resource processing module is used to acquire peak-shaving resource information of each region, discretize the peak-shaving resources and process the non-convexity by using piecewise linear convex envoy relaxation to obtain the discrete level of the peak-shaving resources, and construct a discrete peak-shaving resource pool. The region division module is used to obtain the parameters of the connection lines between the regions, construct a region coupling graph based on the parameters of the connection lines, and divide the regions based on the region coupling graph to obtain multiple scheduling sub-regions; The initial solution acquisition module is used to construct the total system cost function with the discrete gears of the peak-shaving resources in each time period as decision variables, and obtain an initial feasible solution that satisfies the equipment operation constraints. The iterative optimization module is used to obtain candidate exchanges based on the scheduling sub-region, perform constraint verification on the candidate exchanges, calculate the cost change corresponding to the candidate exchanges that pass the verification according to the total system cost function, and iteratively update the initial feasible solution based on the cost change to obtain the current allocation solution. The execution correction module is used to distribute the current allocation solution as a peak-shaving resource allocation scheme to the edge nodes of each region for execution, and to make local corrections to the peak-shaving resource allocation scheme based on the execution feedback information of the edge nodes.

2. The regional power peak-shaving resource allocation system based on cloud computing according to claim 1, characterized in that, The resource processing module is specifically used for: The continuously adjustable resources in the peak-shaving resources are discretized into integer levels according to a preset step size; For the non-convex cost curve of the peak-shaving resource, a piecewise linear convex envelope relaxation model is constructed, and the non-convexity is handled in the objective function; Based on the processed peak-shaving resources, the corresponding integer levels, and the relaxed convex hull cost curve, a discrete peak-shaving resource pool is formed.

3. The regional power peak-shaving resource allocation system based on cloud computing according to claim 1, characterized in that, When constructing a region coupling diagram based on the parameters of the connecting lines, the region partitioning module is specifically used for: Obtain the equivalent reactance and transmission capacity of the tie line; The initial electrical coupling weight of the tie line is determined based on the equivalent reactance and the transmission capacity, and the initial electrical coupling weight is adjusted in combination with historical peak-shaving resource mutual assistance direction data to obtain the final coupling weight. A regional coupling graph is constructed using each region as a node and the connecting lines with the coupling weights as edges.

4. The regional power peak-shaving resource allocation system based on cloud computing according to claim 3, characterized in that, When the region partitioning module divides the regions based on the region coupling graph to obtain multiple scheduling sub-regions, it is specifically used for: Calculate the Laplacian matrix corresponding to the region coupling graph; Find the eigenvector corresponding to the second smallest eigenvalue in the Laplacian matrix; The regions are recursively divided into two parts according to the positive or negative signs of the elements in the feature vector, until the number of regions contained in each sub-region after division meets the preset condition, thus obtaining multiple scheduling sub-regions.

5. The regional power peak-shaving resource allocation system based on cloud computing according to claim 1, characterized in that, The total system cost function includes a resource allocation cost item, a cross-regional imbalance penalty item, a insufficient peak shaving penalty item, and an excessive peak shaving penalty item; The resource allocation cost item adopts a discrete convex function form, the cross-regional imbalance penalty item adopts a discrete convex function applied to the difference in the total peak-shaving power of different regions in each region, and the peak-shaving insufficient penalty item and the peak-shaving excess penalty item adopt discrete convex functions applied to the deviation between the net load and the total peak-shaving output of each region, respectively. The total cost function of the system satisfies the M-convex property over the integer feasible region of the decision variables.

6. The regional power peak-shaving resource allocation system based on cloud computing according to claim 5, characterized in that, The formula for calculating the total system cost function is as follows: ； In the formula: The total system scheduling cost; This refers to the discrete levels of peak-shaving resources at different times. This is a secondary cost coefficient. This is the primary cost coefficient; and These represent the total power already utilized by the two areas with interconnecting lines. For the coupling weights of the corresponding tie lines, This is the penalty coefficient for cross-regional imbalance; This represents the power deviation between net load and total peak-shaving output. This is the deviation penalty coefficient.

7. The regional power peak-shaving resource allocation system based on cloud computing according to claim 1, characterized in that, When the iterative optimization module obtains candidate swaps based on the scheduling sub-region, it is specifically used for: Prioritize obtaining the first candidate swap within any of the aforementioned scheduling sub-regions; If there is no first candidate exchange within the scheduling sub-region that reduces the total cost function of the system, then cross-regional exchange is used as a secondary backup strategy to obtain a second candidate exchange between different scheduling sub-regions. Both the first candidate exchange and the second candidate exchange are discrete gear shifts between the two peak-shaving resources.

8. The regional power peak-shaving resource allocation system based on cloud computing according to claim 1, characterized in that, The constraints on device operation and the constraints on candidate exchanges both include upper and lower limit constraints on resources, ramp rate constraints, and tie line capacity constraints.

9. The regional power peak-shaving resource allocation system based on cloud computing according to claim 1, characterized in that, It also includes a scene prediction module, used for: Obtain net load prediction error data, and select multiple quantile levels based on the distribution of the net load prediction error data to form a typical net load scenario set containing scenario probabilities; The typical net load scenario set is used to calculate the expected cost for multiple scenarios when constructing the total cost function of the system.

10. The regional power peak-shaving resource allocation system based on cloud computing according to claim 1, characterized in that, When the execution correction module performs local corrections to the peak-shaving resource allocation scheme based on the execution feedback information from the edge nodes, it is specifically used for: After receiving the edge node's local power flow security verification and execution of the peak shaving resource allocation scheme, the uploaded actual execution level and deviation reason are used as the execution feedback information. Based on the execution feedback information, the affected area is determined. While keeping the allocation scheme of other areas unchanged, a local candidate exchange is obtained within the affected area to update the peak-shaving resource allocation scheme.

Images